first

.gitattributes

@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,9 @@
+/.idea/vcs.xml
+/hubert/put_hubert_ckpt_here
+/.idea/inspectionProfiles/Project_Default.xml
+/.idea/inspectionProfiles/profiles_settings.xml
+/.idea/modules.xml
+/.idea/.gitignore
+/.idea/so-vits-svc-3.0-app.iml
+/.idea/AI-minato_aqua.iml
+/.idea/misc.xml

app.py

@@ -0,0 +1,76 @@
+import io
+import os
+
+import gradio as gr
+import librosa
+import numpy as np
+import soundfile
+import torch
+from inference.infer_tool import Svc
+import logging
+logging.getLogger("numba").setLevel(logging.WARNING)
+
+model_path = "logs/48k/aqua.pth"
+config_path = "configs/config.json"
+
+svc_model = Svc(model_path, config_path)
+
+
+def vc_fn(input_audio, vc_transform, term):
+    if not term:
+        return "请阅读并同意《AI阿夸模型使用协议》", None
+    if input_audio is None:
+        return "请上传音频", None
+    sampling_rate, audio = input_audio
+    duration = audio.shape[0] / sampling_rate
+    if duration > 30:
+        return "请上传小于30s的音频，长音频的转换请在本地进行", None
+    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
+    if len(audio.shape) > 1:
+        audio = librosa.to_mono(audio.transpose(1, 0))
+    if sampling_rate != 24000:
+        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=24000)
+    print(audio.shape)
+    out_wav_path = io.BytesIO()
+    soundfile.write(out_wav_path, audio, 24000, format="wav")
+    out_wav_path.seek(0)
+
+    # sid = sid_map[sid]
+    sid = "aqua"
+    out_audio, out_sr = svc_model.infer(sid, vc_transform, out_wav_path)
+    _audio = out_audio.cpu().numpy()
+    return "Success", (48000, _audio)
+
+
+inputs = [
+    gr.inputs.Audio(source="upload"),
+    gr.inputs.Number(default=0),
+    gr.Checkbox(label="您已阅读并同意《AI阿夸模型使用协议》")
+]
+
+outputs = [
+    "text",
+    gr.outputs.Audio(type="numpy")
+]
+
+example = [
+    ["./raw/大手拉小手.wav", 0, False]
+]
+
+des = """
+## 在使用此模型前请阅读[AI阿夸模型使用协议](https://huggingface.co/spaces/DoNotSelect/AI-minato_aqua/blob/main/terms.md)
+"""
+
+demo = gr.Interface(
+    fn=vc_fn,
+    inputs=inputs,
+    outputs=outputs,
+    layout="horizontal",
+    theme="huggingface",
+    description=des,
+    examples=example,
+    cache_examples=True
+)
+
+if __name__ == "__main__":
+    demo.launch()

attentions.py

@@ -0,0 +1,303 @@
+import copy
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import modules
+from modules import LayerNorm
+   
+
+class Encoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., window_size=4, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+
+    self.drop = nn.Dropout(p_dropout)
+    self.attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, window_size=window_size))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask):
+    attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.attn_layers[i](x, x, attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class Decoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., proximal_bias=False, proximal_init=True, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+
+    self.drop = nn.Dropout(p_dropout)
+    self.self_attn_layers = nn.ModuleList()
+    self.norm_layers_0 = nn.ModuleList()
+    self.encdec_attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.self_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, proximal_bias=proximal_bias, proximal_init=proximal_init))
+      self.norm_layers_0.append(LayerNorm(hidden_channels))
+      self.encdec_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout, causal=True))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask, h, h_mask):
+    """
+    x: decoder input
+    h: encoder output
+    """
+    self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(device=x.device, dtype=x.dtype)
+    encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.self_attn_layers[i](x, x, self_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_0[i](x + y)
+
+      y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+      
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class MultiHeadAttention(nn.Module):
+  def __init__(self, channels, out_channels, n_heads, p_dropout=0., window_size=None, heads_share=True, block_length=None, proximal_bias=False, proximal_init=False):
+    super().__init__()
+    assert channels % n_heads == 0
+
+    self.channels = channels
+    self.out_channels = out_channels
+    self.n_heads = n_heads
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+    self.heads_share = heads_share
+    self.block_length = block_length
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+    self.attn = None
+
+    self.k_channels = channels // n_heads
+    self.conv_q = nn.Conv1d(channels, channels, 1)
+    self.conv_k = nn.Conv1d(channels, channels, 1)
+    self.conv_v = nn.Conv1d(channels, channels, 1)
+    self.conv_o = nn.Conv1d(channels, out_channels, 1)
+    self.drop = nn.Dropout(p_dropout)
+
+    if window_size is not None:
+      n_heads_rel = 1 if heads_share else n_heads
+      rel_stddev = self.k_channels**-0.5
+      self.emb_rel_k = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+      self.emb_rel_v = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+
+    nn.init.xavier_uniform_(self.conv_q.weight)
+    nn.init.xavier_uniform_(self.conv_k.weight)
+    nn.init.xavier_uniform_(self.conv_v.weight)
+    if proximal_init:
+      with torch.no_grad():
+        self.conv_k.weight.copy_(self.conv_q.weight)
+        self.conv_k.bias.copy_(self.conv_q.bias)
+      
+  def forward(self, x, c, attn_mask=None):
+    q = self.conv_q(x)
+    k = self.conv_k(c)
+    v = self.conv_v(c)
+    
+    x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+    x = self.conv_o(x)
+    return x
+
+  def attention(self, query, key, value, mask=None):
+    # reshape [b, d, t] -> [b, n_h, t, d_k]
+    b, d, t_s, t_t = (*key.size(), query.size(2))
+    query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+    key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+    value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+    scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+    if self.window_size is not None:
+      assert t_s == t_t, "Relative attention is only available for self-attention."
+      key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+      rel_logits = self._matmul_with_relative_keys(query /math.sqrt(self.k_channels), key_relative_embeddings)
+      scores_local = self._relative_position_to_absolute_position(rel_logits)
+      scores = scores + scores_local
+    if self.proximal_bias:
+      assert t_s == t_t, "Proximal bias is only available for self-attention."
+      scores = scores + self._attention_bias_proximal(t_s).to(device=scores.device, dtype=scores.dtype)
+    if mask is not None:
+      scores = scores.masked_fill(mask == 0, -1e4)
+      if self.block_length is not None:
+        assert t_s == t_t, "Local attention is only available for self-attention."
+        block_mask = torch.ones_like(scores).triu(-self.block_length).tril(self.block_length)
+        scores = scores.masked_fill(block_mask == 0, -1e4)
+    p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
+    p_attn = self.drop(p_attn)
+    output = torch.matmul(p_attn, value)
+    if self.window_size is not None:
+      relative_weights = self._absolute_position_to_relative_position(p_attn)
+      value_relative_embeddings = self._get_relative_embeddings(self.emb_rel_v, t_s)
+      output = output + self._matmul_with_relative_values(relative_weights, value_relative_embeddings)
+    output = output.transpose(2, 3).contiguous().view(b, d, t_t) # [b, n_h, t_t, d_k] -> [b, d, t_t]
+    return output, p_attn
+
+  def _matmul_with_relative_values(self, x, y):
+    """
+    x: [b, h, l, m]
+    y: [h or 1, m, d]
+    ret: [b, h, l, d]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0))
+    return ret
+
+  def _matmul_with_relative_keys(self, x, y):
+    """
+    x: [b, h, l, d]
+    y: [h or 1, m, d]
+    ret: [b, h, l, m]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+    return ret
+
+  def _get_relative_embeddings(self, relative_embeddings, length):
+    max_relative_position = 2 * self.window_size + 1
+    # Pad first before slice to avoid using cond ops.
+    pad_length = max(length - (self.window_size + 1), 0)
+    slice_start_position = max((self.window_size + 1) - length, 0)
+    slice_end_position = slice_start_position + 2 * length - 1
+    if pad_length > 0:
+      padded_relative_embeddings = F.pad(
+          relative_embeddings,
+          commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]))
+    else:
+      padded_relative_embeddings = relative_embeddings
+    used_relative_embeddings = padded_relative_embeddings[:,slice_start_position:slice_end_position]
+    return used_relative_embeddings
+
+  def _relative_position_to_absolute_position(self, x):
+    """
+    x: [b, h, l, 2*l-1]
+    ret: [b, h, l, l]
+    """
+    batch, heads, length, _ = x.size()
+    # Concat columns of pad to shift from relative to absolute indexing.
+    x = F.pad(x, commons.convert_pad_shape([[0,0],[0,0],[0,0],[0,1]]))
+
+    # Concat extra elements so to add up to shape (len+1, 2*len-1).
+    x_flat = x.view([batch, heads, length * 2 * length])
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0,0],[0,0],[0,length-1]]))
+
+    # Reshape and slice out the padded elements.
+    x_final = x_flat.view([batch, heads, length+1, 2*length-1])[:, :, :length, length-1:]
+    return x_final
+
+  def _absolute_position_to_relative_position(self, x):
+    """
+    x: [b, h, l, l]
+    ret: [b, h, l, 2*l-1]
+    """
+    batch, heads, length, _ = x.size()
+    # padd along column
+    x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length-1]]))
+    x_flat = x.view([batch, heads, length**2 + length*(length -1)])
+    # add 0's in the beginning that will skew the elements after reshape
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+    x_final = x_flat.view([batch, heads, length, 2*length])[:,:,:,1:]
+    return x_final
+
+  def _attention_bias_proximal(self, length):
+    """Bias for self-attention to encourage attention to close positions.
+    Args:
+      length: an integer scalar.
+    Returns:
+      a Tensor with shape [1, 1, length, length]
+    """
+    r = torch.arange(length, dtype=torch.float32)
+    diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+    return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+  def __init__(self, in_channels, out_channels, filter_channels, kernel_size, p_dropout=0., activation=None, causal=False):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.activation = activation
+    self.causal = causal
+
+    if causal:
+      self.padding = self._causal_padding
+    else:
+      self.padding = self._same_padding
+
+    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+    self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+    self.drop = nn.Dropout(p_dropout)
+
+  def forward(self, x, x_mask):
+    x = self.conv_1(self.padding(x * x_mask))
+    if self.activation == "gelu":
+      x = x * torch.sigmoid(1.702 * x)
+    else:
+      x = torch.relu(x)
+    x = self.drop(x)
+    x = self.conv_2(self.padding(x * x_mask))
+    return x * x_mask
+  
+  def _causal_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = self.kernel_size - 1
+    pad_r = 0
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x
+
+  def _same_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = (self.kernel_size - 1) // 2
+    pad_r = self.kernel_size // 2
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x

commons.py

@@ -0,0 +1,188 @@
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+def slice_pitch_segments(x, ids_str, segment_size=4):
+  ret = torch.zeros_like(x[:, :segment_size])
+  for i in range(x.size(0)):
+    idx_str = ids_str[i]
+    idx_end = idx_str + segment_size
+    ret[i] = x[i, idx_str:idx_end]
+  return ret
+
+def rand_slice_segments_with_pitch(x, pitch, x_lengths=None, segment_size=4):
+  b, d, t = x.size()
+  if x_lengths is None:
+    x_lengths = t
+  ids_str_max = x_lengths - segment_size + 1
+  ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+  ret = slice_segments(x, ids_str, segment_size)
+  ret_pitch = slice_pitch_segments(pitch, ids_str, segment_size)
+  return ret, ret_pitch, ids_str
+
+def init_weights(m, mean=0.0, std=0.01):
+  classname = m.__class__.__name__
+  if classname.find("Conv") != -1:
+    m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+  return int((kernel_size*dilation - dilation)/2)
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def intersperse(lst, item):
+  result = [item] * (len(lst) * 2 + 1)
+  result[1::2] = lst
+  return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+  """KL(P||Q)"""
+  kl = (logs_q - logs_p) - 0.5
+  kl += 0.5 * (torch.exp(2. * logs_p) + ((m_p - m_q)**2)) * torch.exp(-2. * logs_q)
+  return kl
+
+
+def rand_gumbel(shape):
+  """Sample from the Gumbel distribution, protect from overflows."""
+  uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+  return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+  g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+  return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+  ret = torch.zeros_like(x[:, :, :segment_size])
+  for i in range(x.size(0)):
+    idx_str = ids_str[i]
+    idx_end = idx_str + segment_size
+    ret[i] = x[i, :, idx_str:idx_end]
+  return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+  b, d, t = x.size()
+  if x_lengths is None:
+    x_lengths = t
+  ids_str_max = x_lengths - segment_size + 1
+  ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+  ret = slice_segments(x, ids_str, segment_size)
+  return ret, ids_str
+
+
+def rand_spec_segments(x, x_lengths=None, segment_size=4):
+  b, d, t = x.size()
+  if x_lengths is None:
+    x_lengths = t
+  ids_str_max = x_lengths - segment_size
+  ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+  ret = slice_segments(x, ids_str, segment_size)
+  return ret, ids_str
+
+
+def get_timing_signal_1d(
+    length, channels, min_timescale=1.0, max_timescale=1.0e4):
+  position = torch.arange(length, dtype=torch.float)
+  num_timescales = channels // 2
+  log_timescale_increment = (
+      math.log(float(max_timescale) / float(min_timescale)) /
+      (num_timescales - 1))
+  inv_timescales = min_timescale * torch.exp(
+      torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment)
+  scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+  signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+  signal = F.pad(signal, [0, 0, 0, channels % 2])
+  signal = signal.view(1, channels, length)
+  return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+  mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+  return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+  n_channels_int = n_channels[0]
+  in_act = input_a + input_b
+  t_act = torch.tanh(in_act[:, :n_channels_int, :])
+  s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+  acts = t_act * s_act
+  return acts
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def shift_1d(x):
+  x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+  return x
+
+
+def sequence_mask(length, max_length=None):
+  if max_length is None:
+    max_length = length.max()
+  x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+  return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+  """
+  duration: [b, 1, t_x]
+  mask: [b, 1, t_y, t_x]
+  """
+  device = duration.device
+  
+  b, _, t_y, t_x = mask.shape
+  cum_duration = torch.cumsum(duration, -1)
+  
+  cum_duration_flat = cum_duration.view(b * t_x)
+  path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+  path = path.view(b, t_x, t_y)
+  path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+  path = path.unsqueeze(1).transpose(2,3) * mask
+  return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+  if isinstance(parameters, torch.Tensor):
+    parameters = [parameters]
+  parameters = list(filter(lambda p: p.grad is not None, parameters))
+  norm_type = float(norm_type)
+  if clip_value is not None:
+    clip_value = float(clip_value)
+
+  total_norm = 0
+  for p in parameters:
+    param_norm = p.grad.data.norm(norm_type)
+    total_norm += param_norm.item() ** norm_type
+    if clip_value is not None:
+      p.grad.data.clamp_(min=-clip_value, max=clip_value)
+  total_norm = total_norm ** (1. / norm_type)
+  return total_norm

configs/config.json

@@ -0,0 +1,90 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 10000,
+    "learning_rate": 0.0002,
+    "betas": [
+      0.8,
+      0.99
+    ],
+    "eps": 1e-09,
+    "batch_size": 12,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 17920,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0,
+    "use_sr": true,
+    "max_speclen": 384,
+    "port": "8001"
+  },
+  "data": {
+    "training_files": "filelists/train.txt",
+    "validation_files": "filelists/val.txt",
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 1280,
+    "hop_length": 320,
+    "win_length": 1280,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [
+      3,
+      7,
+      11
+    ],
+    "resblock_dilation_sizes": [
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ]
+    ],
+    "upsample_rates": [
+      10,
+      8,
+      2,
+      2
+    ],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [
+      16,
+      16,
+      4,
+      4
+    ],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "ssl_dim": 256,
+    "n_speakers": 2
+  },
+  "spk": {
+    "aqua": 0
+  }
+}

data_utils.py

@@ -0,0 +1,152 @@
+import time
+import os
+import random
+import numpy as np
+import torch
+import torch.utils.data
+
+import commons
+from mel_processing import spectrogram_torch, spec_to_mel_torch
+from utils import load_wav_to_torch, load_filepaths_and_text, transform
+
+# import h5py
+
+
+"""Multi speaker version"""
+
+
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, speaker_id, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths, hparams):
+        self.audiopaths = load_filepaths_and_text(audiopaths)
+        self.max_wav_value = hparams.data.max_wav_value
+        self.sampling_rate = hparams.data.sampling_rate
+        self.filter_length = hparams.data.filter_length
+        self.hop_length = hparams.data.hop_length
+        self.win_length = hparams.data.win_length
+        self.sampling_rate = hparams.data.sampling_rate
+        self.use_sr = hparams.train.use_sr
+        self.spec_len = hparams.train.max_speclen
+        self.spk_map = hparams.spk
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                                     self.sampling_rate, self.hop_length, self.win_length,
+                                     center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+
+        spk = filename.split(os.sep)[-2]
+        spk = torch.LongTensor([self.spk_map[spk]])
+
+        c = torch.load(filename + ".soft.pt").squeeze(0)
+        c = torch.repeat_interleave(c, repeats=3, dim=1)
+
+        f0 = np.load(filename + ".f0.npy")
+        f0 = torch.FloatTensor(f0)
+        lmin = min(c.size(-1), spec.size(-1), f0.shape[0])
+        assert abs(c.size(-1) - spec.size(-1)) < 4, (c.size(-1), spec.size(-1), f0.shape, filename)
+        assert abs(lmin - spec.size(-1)) < 4, (c.size(-1), spec.size(-1), f0.shape)
+        assert abs(lmin - c.size(-1)) < 4, (c.size(-1), spec.size(-1), f0.shape)
+        spec, c, f0 = spec[:, :lmin], c[:, :lmin], f0[:lmin]
+        audio_norm = audio_norm[:, :lmin * self.hop_length]
+        _spec, _c, _audio_norm, _f0 = spec, c, audio_norm, f0
+        while spec.size(-1) < self.spec_len:
+            spec = torch.cat((spec, _spec), -1)
+            c = torch.cat((c, _c), -1)
+            f0 = torch.cat((f0, _f0), -1)
+            audio_norm = torch.cat((audio_norm, _audio_norm), -1)
+        start = random.randint(0, spec.size(-1) - self.spec_len)
+        end = start + self.spec_len
+        spec = spec[:, start:end]
+        c = c[:, start:end]
+        f0 = f0[start:end]
+        audio_norm = audio_norm[:, start * self.hop_length:end * self.hop_length]
+
+        return c, f0, spec, audio_norm, spk
+
+    def __getitem__(self, index):
+        return self.get_audio(self.audiopaths[index][0])
+
+    def __len__(self):
+        return len(self.audiopaths)
+
+
+class EvalDataLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, speaker_id, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths, hparams):
+        self.audiopaths = load_filepaths_and_text(audiopaths)
+        self.max_wav_value = hparams.data.max_wav_value
+        self.sampling_rate = hparams.data.sampling_rate
+        self.filter_length = hparams.data.filter_length
+        self.hop_length = hparams.data.hop_length
+        self.win_length = hparams.data.win_length
+        self.sampling_rate = hparams.data.sampling_rate
+        self.use_sr = hparams.train.use_sr
+        self.audiopaths = self.audiopaths[:5]
+        self.spk_map = hparams.spk
+
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                                     self.sampling_rate, self.hop_length, self.win_length,
+                                     center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+
+        spk = filename.split(os.sep)[-2]
+        spk = torch.LongTensor([self.spk_map[spk]])
+
+        c = torch.load(filename + ".soft.pt").squeeze(0)
+
+        c = torch.repeat_interleave(c, repeats=3, dim=1)
+
+        f0 = np.load(filename + ".f0.npy")
+        f0 = torch.FloatTensor(f0)
+        lmin = min(c.size(-1), spec.size(-1), f0.shape[0])
+        assert abs(c.size(-1) - spec.size(-1)) < 4, (c.size(-1), spec.size(-1), f0.shape)
+        assert abs(f0.shape[0] - spec.shape[-1]) < 4, (c.size(-1), spec.size(-1), f0.shape)
+        spec, c, f0 = spec[:, :lmin], c[:, :lmin], f0[:lmin]
+        audio_norm = audio_norm[:, :lmin * self.hop_length]
+
+        return c, f0, spec, audio_norm, spk
+
+    def __getitem__(self, index):
+        return self.get_audio(self.audiopaths[index][0])
+
+    def __len__(self):
+        return len(self.audiopaths)
+

flask_api.py

@@ -0,0 +1,56 @@
+import io
+import logging
+
+import soundfile
+import torch
+import torchaudio
+from flask import Flask, request, send_file
+from flask_cors import CORS
+
+from inference.infer_tool import Svc, RealTimeVC
+
+app = Flask(__name__)
+
+CORS(app)
+
+logging.getLogger('numba').setLevel(logging.WARNING)
+
+
+@app.route("/voiceChangeModel", methods=["POST"])
+def voice_change_model():
+    request_form = request.form
+    wave_file = request.files.get("sample", None)
+    # 变调信息
+    f_pitch_change = float(request_form.get("fPitchChange", 0))
+    # DAW所需的采样率
+    daw_sample = int(float(request_form.get("sampleRate", 0)))
+    speaker_id = int(float(request_form.get("sSpeakId", 0)))
+    # http获得wav文件并转换
+    input_wav_path = io.BytesIO(wave_file.read())
+
+    # 模型推理
+    if raw_infer:
+        out_audio, out_sr = svc_model.infer(speaker_id, f_pitch_change, input_wav_path)
+        tar_audio = torchaudio.functional.resample(out_audio, svc_model.target_sample, daw_sample)
+    else:
+        out_audio = svc.process(svc_model, speaker_id, f_pitch_change, input_wav_path)
+        tar_audio = torchaudio.functional.resample(torch.from_numpy(out_audio), svc_model.target_sample, daw_sample)
+    # 返回音频
+    out_wav_path = io.BytesIO()
+    soundfile.write(out_wav_path, tar_audio.cpu().numpy(), daw_sample, format="wav")
+    out_wav_path.seek(0)
+    return send_file(out_wav_path, download_name="temp.wav", as_attachment=True)
+
+
+if __name__ == '__main__':
+    # 启用则为直接切片合成，False为交叉淡化方式
+    # vst插件调整0.3-0.5s切片时间可以降低延迟，直接切片方法会有连接处爆音、交叉淡化会有轻微重叠声音
+    # 自行选择能接受的方法，或将vst最大切片时间调整为1s，此处设为Ture，延迟大音质稳定一些
+    raw_infer = True
+    # 每个模型和config是唯一对应的
+    model_name = "logs/48k/G_174000-Copy1.pth"
+    config_name = "configs/config.json"
+    svc_model = Svc(model_name, config_name)
+    svc = RealTimeVC()
+    # 此处与vst插件对应，不建议更改
+    app.run(port=6842, host="0.0.0.0", debug=False, threaded=False)

hubert/hubert-soft-0d54a1f4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e82e7d079df05fe3aa535f6f7d42d309bdae1d2a53324e2b2386c56721f4f649
+size 378435957

hubert/hubert_model.py

@@ -0,0 +1,222 @@
+import copy
+import random
+from typing import Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as t_func
+from torch.nn.modules.utils import consume_prefix_in_state_dict_if_present
+
+
+class Hubert(nn.Module):
+    def __init__(self, num_label_embeddings: int = 100, mask: bool = True):
+        super().__init__()
+        self._mask = mask
+        self.feature_extractor = FeatureExtractor()
+        self.feature_projection = FeatureProjection()
+        self.positional_embedding = PositionalConvEmbedding()
+        self.norm = nn.LayerNorm(768)
+        self.dropout = nn.Dropout(0.1)
+        self.encoder = TransformerEncoder(
+            nn.TransformerEncoderLayer(
+                768, 12, 3072, activation="gelu", batch_first=True
+            ),
+            12,
+        )
+        self.proj = nn.Linear(768, 256)
+
+        self.masked_spec_embed = nn.Parameter(torch.FloatTensor(768).uniform_())
+        self.label_embedding = nn.Embedding(num_label_embeddings, 256)
+
+    def mask(self, x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        mask = None
+        if self.training and self._mask:
+            mask = _compute_mask((x.size(0), x.size(1)), 0.8, 10, x.device, 2)
+            x[mask] = self.masked_spec_embed.to(x.dtype)
+        return x, mask
+
+    def encode(
+            self, x: torch.Tensor, layer: Optional[int] = None
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        x = self.feature_extractor(x)
+        x = self.feature_projection(x.transpose(1, 2))
+        x, mask = self.mask(x)
+        x = x + self.positional_embedding(x)
+        x = self.dropout(self.norm(x))
+        x = self.encoder(x, output_layer=layer)
+        return x, mask
+
+    def logits(self, x: torch.Tensor) -> torch.Tensor:
+        logits = torch.cosine_similarity(
+            x.unsqueeze(2),
+            self.label_embedding.weight.unsqueeze(0).unsqueeze(0),
+            dim=-1,
+        )
+        return logits / 0.1
+
+    def forward(self, x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        x, mask = self.encode(x)
+        x = self.proj(x)
+        logits = self.logits(x)
+        return logits, mask
+
+
+class HubertSoft(Hubert):
+    def __init__(self):
+        super().__init__()
+
+    @torch.inference_mode()
+    def units(self, wav: torch.Tensor) -> torch.Tensor:
+        wav = t_func.pad(wav, ((400 - 320) // 2, (400 - 320) // 2))
+        x, _ = self.encode(wav)
+        return self.proj(x)
+
+
+class FeatureExtractor(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.conv0 = nn.Conv1d(1, 512, 10, 5, bias=False)
+        self.norm0 = nn.GroupNorm(512, 512)
+        self.conv1 = nn.Conv1d(512, 512, 3, 2, bias=False)
+        self.conv2 = nn.Conv1d(512, 512, 3, 2, bias=False)
+        self.conv3 = nn.Conv1d(512, 512, 3, 2, bias=False)
+        self.conv4 = nn.Conv1d(512, 512, 3, 2, bias=False)
+        self.conv5 = nn.Conv1d(512, 512, 2, 2, bias=False)
+        self.conv6 = nn.Conv1d(512, 512, 2, 2, bias=False)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = t_func.gelu(self.norm0(self.conv0(x)))
+        x = t_func.gelu(self.conv1(x))
+        x = t_func.gelu(self.conv2(x))
+        x = t_func.gelu(self.conv3(x))
+        x = t_func.gelu(self.conv4(x))
+        x = t_func.gelu(self.conv5(x))
+        x = t_func.gelu(self.conv6(x))
+        return x
+
+
+class FeatureProjection(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.norm = nn.LayerNorm(512)
+        self.projection = nn.Linear(512, 768)
+        self.dropout = nn.Dropout(0.1)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.norm(x)
+        x = self.projection(x)
+        x = self.dropout(x)
+        return x
+
+
+class PositionalConvEmbedding(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.conv = nn.Conv1d(
+            768,
+            768,
+            kernel_size=128,
+            padding=128 // 2,
+            groups=16,
+        )
+        self.conv = nn.utils.weight_norm(self.conv, name="weight", dim=2)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.conv(x.transpose(1, 2))
+        x = t_func.gelu(x[:, :, :-1])
+        return x.transpose(1, 2)
+
+
+class TransformerEncoder(nn.Module):
+    def __init__(
+            self, encoder_layer: nn.TransformerEncoderLayer, num_layers: int
+    ) -> None:
+        super(TransformerEncoder, self).__init__()
+        self.layers = nn.ModuleList(
+            [copy.deepcopy(encoder_layer) for _ in range(num_layers)]
+        )
+        self.num_layers = num_layers
+
+    def forward(
+            self,
+            src: torch.Tensor,
+            mask: torch.Tensor = None,
+            src_key_padding_mask: torch.Tensor = None,
+            output_layer: Optional[int] = None,
+    ) -> torch.Tensor:
+        output = src
+        for layer in self.layers[:output_layer]:
+            output = layer(
+                output, src_mask=mask, src_key_padding_mask=src_key_padding_mask
+            )
+        return output
+
+
+def _compute_mask(
+        shape: Tuple[int, int],
+        mask_prob: float,
+        mask_length: int,
+        device: torch.device,
+        min_masks: int = 0,
+) -> torch.Tensor:
+    batch_size, sequence_length = shape
+
+    if mask_length < 1:
+        raise ValueError("`mask_length` has to be bigger than 0.")
+
+    if mask_length > sequence_length:
+        raise ValueError(
+            f"`mask_length` has to be smaller than `sequence_length`, but got `mask_length`: {mask_length} and `sequence_length`: {sequence_length}`"
+        )
+
+    # compute number of masked spans in batch
+    num_masked_spans = int(mask_prob * sequence_length / mask_length + random.random())
+    num_masked_spans = max(num_masked_spans, min_masks)
+
+    # make sure num masked indices <= sequence_length
+    if num_masked_spans * mask_length > sequence_length:
+        num_masked_spans = sequence_length // mask_length
+
+    # SpecAugment mask to fill
+    mask = torch.zeros((batch_size, sequence_length), device=device, dtype=torch.bool)
+
+    # uniform distribution to sample from, make sure that offset samples are < sequence_length
+    uniform_dist = torch.ones(
+        (batch_size, sequence_length - (mask_length - 1)), device=device
+    )
+
+    # get random indices to mask
+    mask_indices = torch.multinomial(uniform_dist, num_masked_spans)
+
+    # expand masked indices to masked spans
+    mask_indices = (
+        mask_indices.unsqueeze(dim=-1)
+        .expand((batch_size, num_masked_spans, mask_length))
+        .reshape(batch_size, num_masked_spans * mask_length)
+    )
+    offsets = (
+        torch.arange(mask_length, device=device)[None, None, :]
+        .expand((batch_size, num_masked_spans, mask_length))
+        .reshape(batch_size, num_masked_spans * mask_length)
+    )
+    mask_idxs = mask_indices + offsets
+
+    # scatter indices to mask
+    mask = mask.scatter(1, mask_idxs, True)
+
+    return mask
+
+
+def hubert_soft(
+        path: str,
+) -> HubertSoft:
+    r"""HuBERT-Soft from `"A Comparison of Discrete and Soft Speech Units for Improved Voice Conversion"`.
+    Args:
+        path (str): path of a pretrained model
+    """
+    hubert = HubertSoft()
+    checkpoint = torch.load(path)
+    consume_prefix_in_state_dict_if_present(checkpoint, "module.")
+    hubert.load_state_dict(checkpoint)
+    hubert.eval()
+    return hubert

inference/chunks_temp.json

@@ -0,0 +1 @@
+{"6da25b17631773f3e267d9ea5e90fdf2": {"chunks": {"0": {"slice": true, "split_time": "0,23835"}, "1": {"slice": false, "split_time": "23835,905348"}, "2": {"slice": true, "split_time": "905348,910805"}, "3": {"slice": false, "split_time": "910805,1395266"}, "4": {"slice": true, "split_time": "1395266,1398353"}, "5": {"slice": false, "split_time": "1398353,2637030"}, "6": {"slice": true, "split_time": "2637030,2657548"}, "7": {"slice": false, "split_time": "2657548,3360683"}, "8": {"slice": true, "split_time": "3360683,3363848"}, "9": {"slice": false, "split_time": "3363848,4478614"}, "10": {"slice": true, "split_time": "4478614,4514816"}}, "time": 1670571633}, "90b894a25dfbca42c065e3a27d75a1b7": {"chunks": {"0": {"slice": true, "split_time": "0,24809"}, "1": {"slice": false, "split_time": "24809,905116"}, "2": {"slice": true, "split_time": "905116,910645"}, "3": {"slice": false, "split_time": "910645,1234800"}}, "time": 1670569815}, "c011425d4cb059beeef217ea9c291f67": {"chunks": {"0": {"slice": true, "split_time": "0,4304"}, "1": {"slice": false, "split_time": "4304,335402"}, "2": {"slice": true, "split_time": "335402,643324"}, "3": {"slice": false, "split_time": "643324,1659483"}}, "time": 1670570441}, "b8225c3a47ffea84e7a9639fde444c52": {"chunks": {"0": {"slice": true, "split_time": "0,61578"}, "1": {"slice": false, "split_time": "61578,1970306"}, "2": {"slice": true, "split_time": "1970306,1982351"}, "3": {"slice": false, "split_time": "1982351,3037480"}, "4": {"slice": true, "split_time": "3037480,3044033"}, "5": {"slice": false, "split_time": "3044033,5739816"}, "6": {"slice": true, "split_time": "5739816,5784949"}, "7": {"slice": false, "split_time": "5784949,7315823"}, "8": {"slice": true, "split_time": "7315823,7322762"}, "9": {"slice": false, "split_time": "7322762,9748878"}, "10": {"slice": true, "split_time": "9748878,9828171"}}, "time": 1670573169}, "36ad08d22d6241f08419ba19448bc1f1": {"chunks": {"0": {"slice": false, "split_time": "0,1136029"}, "1": {"slice": true, "split_time": "1136029,1153805"}, "2": {"slice": false, "split_time": "1153805,3491983"}}, "time": 1670644168}, "b884a288b565236668cc310088504c19": {"chunks": {"0": {"slice": false, "split_time": "0,216521"}, "1": {"slice": true, "split_time": "216521,219197"}, "2": {"slice": false, "split_time": "219197,400384"}, "3": {"slice": true, "split_time": "400384,405224"}, "4": {"slice": false, "split_time": "405224,618976"}, "5": {"slice": true, "split_time": "618976,624194"}, "6": {"slice": false, "split_time": "624194,764242"}, "7": {"slice": true, "split_time": "764242,766510"}, "8": {"slice": false, "split_time": "766510,1433828"}, "9": {"slice": true, "split_time": "1433828,1438002"}, "10": {"slice": false, "split_time": "1438002,1573799"}, "11": {"slice": true, "split_time": "1573799,1576760"}, "12": {"slice": false, "split_time": "1576760,1717339"}, "13": {"slice": true, "split_time": "1717339,1720688"}, "14": {"slice": false, "split_time": "1720688,1891282"}, "15": {"slice": true, "split_time": "1891282,1895616"}, "16": {"slice": false, "split_time": "1895616,2302548"}, "17": {"slice": true, "split_time": "2302548,2305292"}, "18": {"slice": false, "split_time": "2305292,2427056"}, "19": {"slice": true, "split_time": "2427056,2428698"}, "20": {"slice": false, "split_time": "2428698,2541785"}, "21": {"slice": true, "split_time": "2541785,2543993"}, "22": {"slice": false, "split_time": "2543993,2764393"}, "23": {"slice": true, "split_time": "2764393,2771788"}, "24": {"slice": false, "split_time": "2771788,2984762"}, "25": {"slice": true, "split_time": "2984762,2990841"}, "26": {"slice": false, "split_time": "2990841,3324911"}, "27": {"slice": true, "split_time": "3324911,3328212"}, "28": {"slice": false, "split_time": "3328212,3613048"}, "29": {"slice": true, "split_time": "3613048,3613552"}, "30": {"slice": false, "split_time": "3613552,4165908"}, "31": {"slice": true, "split_time": "4165908,4168709"}, "32": {"slice": false, "split_time": "4168709,4220263"}, "33": {"slice": true, "split_time": "4220263,4227485"}}, "time": 1670574749}, "9f02a93ae99d971888b35c785eb7d49b": {"chunks": {"0": {"slice": true, "split_time": "0,705669"}, "1": {"slice": false, "split_time": "705669,6527896"}, "2": {"slice": true, "split_time": "6527896,6597762"}, "3": {"slice": false, "split_time": "6597762,9310902"}, "4": {"slice": true, "split_time": "9310902,9314746"}, "5": {"slice": false, "split_time": "9314746,11243041"}, "6": {"slice": true, "split_time": "11243041,11283279"}, "7": {"slice": false, "split_time": "11283279,11346943"}, "8": {"slice": true, "split_time": "11346943,11630411"}}, "time": 1670575576}, "bf323c2270a30894ddfa25b5d3056d69": {"chunks": {"0": {"slice": true, "split_time": "0,705669"}, "1": {"slice": false, "split_time": "705669,3982080"}}, "time": 1670576110}, "26878ada67625dd6335439521f7ac03c": {"chunks": {"0": {"slice": true, "split_time": "0,30461"}, "1": {"slice": false, "split_time": "30461,1581600"}}, "time": 1670687236}, "3b1da6bfd1773aec9a26891f89a3efc4": {"chunks": {"0": {"slice": true, "split_time": "0,9744"}, "1": {"slice": false, "split_time": "9744,2402304"}}, "time": 1670577430}, "2c59bb37d2ddfc8b2b043edcfcbaa906": {"chunks": {"0": {"slice": true, "split_time": "0,19673"}, "1": {"slice": false, "split_time": "19673,1407635"}, "2": {"slice": true, "split_time": "1407635,1455090"}, "3": {"slice": false, "split_time": "1455090,3997397"}, "4": {"slice": true, "split_time": "3997397,4009124"}, "5": {"slice": false, "split_time": "4009124,6527176"}, "6": {"slice": true, "split_time": "6527176,6587813"}, "7": {"slice": false, "split_time": "6587813,9325038"}, "8": {"slice": true, "split_time": "9325038,9341404"}, "9": {"slice": false, "split_time": "9341404,9593839"}, "10": {"slice": true, "split_time": "9593839,9598960"}, "11": {"slice": false, "split_time": "9598960,9987461"}, "12": {"slice": true, "split_time": "9987461,9994148"}, "13": {"slice": false, "split_time": "9994148,10241257"}, "14": {"slice": true, "split_time": "10241257,10243691"}, "15": {"slice": false, "split_time": "10243691,10532345"}, "16": {"slice": true, "split_time": "10532345,10545150"}, "17": {"slice": false, "split_time": "10545150,10999746"}, "18": {"slice": true, "split_time": "10999746,11012824"}, "19": {"slice": false, "split_time": "11012824,11412797"}, "20": {"slice": true, "split_time": "11412797,11632083"}}, "time": 1670583436}, "ac5cb570b28d1bf43db55d3b55b0da49": {"chunks": {"0": {"slice": true, "split_time": "0,46106"}, "1": {"slice": false, "split_time": "46106,1812107"}, "2": {"slice": true, "split_time": "1812107,1821400"}, "3": {"slice": false, "split_time": "1821400,2684212"}, "4": {"slice": true, "split_time": "2684212,2691807"}, "5": {"slice": false, "split_time": "2691807,5278901"}, "6": {"slice": true, "split_time": "5278901,6171752"}, "7": {"slice": false, "split_time": "6171752,6723937"}, "8": {"slice": true, "split_time": "6723937,6727712"}, "9": {"slice": false, "split_time": "6727712,8961094"}, "10": {"slice": true, "split_time": "8961094,9030528"}}, "time": 1670588363}, "c6bd7a95c90241934030c1730662e7cc": {"chunks": {"0": {"slice": true, "split_time": "0,49204"}, "1": {"slice": false, "split_time": "49204,768993"}, "2": {"slice": true, "split_time": "768993,1419710"}, "3": {"slice": false, "split_time": "1419710,1961254"}, "4": {"slice": true, "split_time": "1961254,2002533"}, "5": {"slice": false, "split_time": "2002533,3400434"}, "6": {"slice": true, "split_time": "3400434,4044462"}, "7": {"slice": false, "split_time": "4044462,4586255"}, "8": {"slice": true, "split_time": "4586255,4627965"}, "9": {"slice": false, "split_time": "4627965,6109298"}, "10": {"slice": true, "split_time": "6109298,6754467"}, "11": {"slice": false, "split_time": "6754467,7126180"}, "12": {"slice": true, "split_time": "7126180,7232446"}, "13": {"slice": false, "split_time": "7232446,7510807"}, "14": {"slice": true, "split_time": "7510807,7579500"}, "15": {"slice": false, "split_time": "7579500,7845805"}, "16": {"slice": true, "split_time": "7845805,8321111"}, "17": {"slice": false, "split_time": "8321111,9008414"}, "18": {"slice": true, "split_time": "9008414,9013254"}, "19": {"slice": false, "split_time": "9013254,9329237"}, "20": {"slice": true, "split_time": "9329237,9719592"}, "21": {"slice": false, "split_time": "9719592,10107581"}, "22": {"slice": true, "split_time": "10107581,10113973"}, "23": {"slice": false, "split_time": "10113973,10350874"}, "24": {"slice": true, "split_time": "10350874,10373826"}, "25": {"slice": false, "split_time": "10373826,10600844"}, "26": {"slice": true, "split_time": "10600844,10616456"}, "27": {"slice": false, "split_time": "10616456,10708106"}, "28": {"slice": true, "split_time": "10708106,10749952"}}, "time": 1670606294}, "4b4f0edb689032a6d91572b9250785e7": {"chunks": {"0": {"slice": true, "split_time": "0,185991"}, "1": {"slice": false, "split_time": "185991,914678"}, "2": {"slice": true, "split_time": "914678,922072"}, "3": {"slice": false, "split_time": "922072,3343526"}, "4": {"slice": true, "split_time": "3343526,3346452"}, "5": {"slice": false, "split_time": "3346452,3597657"}, "6": {"slice": true, "split_time": "3597657,3854141"}}, "time": 1670603989}, "3b32214e143c12a9c2087c6f1a6e0079": {"chunks": {"0": {"slice": true, "split_time": "0,1111906"}, "1": {"slice": false, "split_time": "1111906,1384760"}, "2": {"slice": true, "split_time": "1384760,1431263"}, "3": {"slice": false, "split_time": "1431263,1725453"}, "4": {"slice": true, "split_time": "1725453,1772807"}, "5": {"slice": false, "split_time": "1772807,4592249"}, "6": {"slice": true, "split_time": "4592249,4717056"}}, "time": 1670604680}, "e37af0a83a37929baf38aaf41771f666": {"chunks": {"0": {"slice": true, "split_time": "0,712833"}, "1": {"slice": false, "split_time": "712833,968777"}, "2": {"slice": true, "split_time": "968777,1012890"}, "3": {"slice": false, "split_time": "1012890,1302528"}, "4": {"slice": true, "split_time": "1302528,1372173"}, "5": {"slice": false, "split_time": "1372173,1973196"}, "6": {"slice": true, "split_time": "1973196,1976084"}, "7": {"slice": false, "split_time": "1976084,4088114"}, "8": {"slice": true, "split_time": "4088114,4099072"}}, "time": 1670628295}, "1127f4a167db675427156eff884429f5": {"chunks": {"0": {"slice": true, "split_time": "0,718635"}, "1": {"slice": false, "split_time": "718635,1157581"}, "2": {"slice": true, "split_time": "1157581,1172737"}, "3": {"slice": false, "split_time": "1172737,1705170"}, "4": {"slice": true, "split_time": "1705170,1716523"}, "5": {"slice": false, "split_time": "1716523,1960831"}, "6": {"slice": true, "split_time": "1960831,2038462"}, "7": {"slice": false, "split_time": "2038462,2684694"}, "8": {"slice": true, "split_time": "2684694,2707156"}, "9": {"slice": false, "split_time": "2707156,3037090"}, "10": {"slice": true, "split_time": "3037090,3054814"}, "11": {"slice": false, "split_time": "3054814,4308021"}, "12": {"slice": true, "split_time": "4308021,4326415"}, "13": {"slice": false, "split_time": "4326415,4737386"}, "14": {"slice": true, "split_time": "4737386,5415908"}, "15": {"slice": false, "split_time": "5415908,6049275"}, "16": {"slice": true, "split_time": "6049275,6059272"}, "17": {"slice": false, "split_time": "6059272,6665627"}, "18": {"slice": true, "split_time": "6665627,6678679"}, "19": {"slice": false, "split_time": "6678679,8097473"}, "20": {"slice": true, "split_time": "8097473,8141374"}, "21": {"slice": false, "split_time": "8141374,8897047"}, "22": {"slice": true, "split_time": "8897047,9206103"}, "23": {"slice": false, "split_time": "9206103,9683995"}, "24": {"slice": true, "split_time": "9683995,9699995"}, "25": {"slice": false, "split_time": "9699995,10367622"}, "26": {"slice": true, "split_time": "10367622,10373657"}, "27": {"slice": false, "split_time": "10373657,10914289"}, "28": {"slice": true, "split_time": "10914289,12250463"}, "29": {"slice": false, "split_time": "12250463,12309513"}, "30": {"slice": true, "split_time": "12309513,12788736"}}, "time": 1670658498}, "488951d97170e877bdef769489b09113": {"chunks": {"0": {"slice": true, "split_time": "0,1516829"}, "1": {"slice": false, "split_time": "1516829,3428647"}, "2": {"slice": true, "split_time": "3428647,3434675"}, "3": {"slice": false, "split_time": "3434675,4541897"}, "4": {"slice": true, "split_time": "4541897,4556959"}, "5": {"slice": false, "split_time": "4556959,5514779"}, "6": {"slice": true, "split_time": "5514779,6385599"}, "7": {"slice": false, "split_time": "6385599,8295053"}, "8": {"slice": true, "split_time": "8295053,8312032"}, "9": {"slice": false, "split_time": "8312032,11213869"}, "10": {"slice": true, "split_time": "11213869,11339533"}, "11": {"slice": false, "split_time": "11339533,11810282"}, "12": {"slice": true, "split_time": "11810282,12298240"}}, "time": 1670688385}, "355d47ef6a1b96547e9cca1a0391c85c": {"chunks": {"0": {"slice": true, "split_time": "0,36661"}, "1": {"slice": false, "split_time": "36661,630669"}, "2": {"slice": true, "split_time": "630669,680688"}, "3": {"slice": false, "split_time": "680688,1464986"}, "4": {"slice": true, "split_time": "1464986,1585875"}, "5": {"slice": false, "split_time": "1585875,1843986"}, "6": {"slice": true, "split_time": "1843986,1921120"}, "7": {"slice": false, "split_time": "1921120,2490103"}, "8": {"slice": true, "split_time": "2490103,2536079"}, "9": {"slice": false, "split_time": "2536079,3149312"}, "10": {"slice": true, "split_time": "3149312,3185080"}, "11": {"slice": false, "split_time": "3185080,3709183"}, "12": {"slice": true, "split_time": "3709183,3800117"}, "13": {"slice": false, "split_time": "3800117,4508883"}, "14": {"slice": true, "split_time": "4508883,4582236"}, "15": {"slice": false, "split_time": "4582236,6021956"}}, "time": 1670689010}, "d6c8b7cf8a4245702985ac5be1b854ee": {"chunks": {"0": {"slice": true, "split_time": "0,253821"}, "1": {"slice": false, "split_time": "253821,617754"}, "2": {"slice": true, "split_time": "617754,645464"}, "3": {"slice": false, "split_time": "645464,1772902"}, "4": {"slice": true, "split_time": "1772902,1958678"}, "5": {"slice": false, "split_time": "1958678,4417999"}, "6": {"slice": true, "split_time": "4417999,4487483"}, "7": {"slice": false, "split_time": "4487483,4777645"}, "8": {"slice": true, "split_time": "4777645,4786501"}, "9": {"slice": false, "split_time": "4786501,7052109"}, "10": {"slice": true, "split_time": "7052109,7511077"}, "11": {"slice": false, "split_time": "7511077,9717344"}, "12": {"slice": true, "split_time": "9717344,10189630"}, "13": {"slice": false, "split_time": "10189630,10402183"}, "14": {"slice": true, "split_time": "10402183,10442752"}}, "time": 1670692231}, "06e514729c1002271fcf998844ad5830": {"chunks": {"0": {"slice": true, "split_time": "0,492335"}, "1": {"slice": false, "split_time": "492335,729432"}, "2": {"slice": true, "split_time": "729432,991792"}, "3": {"slice": false, "split_time": "991792,1812379"}, "4": {"slice": true, "split_time": "1812379,1852379"}, "5": {"slice": false, "split_time": "1852379,2572754"}, "6": {"slice": true, "split_time": "2572754,2699840"}, "7": {"slice": false, "split_time": "2699840,2922491"}, "8": {"slice": true, "split_time": "2922491,3000707"}, "9": {"slice": false, "split_time": "3000707,3533124"}, "10": {"slice": true, "split_time": "3533124,3575117"}, "11": {"slice": false, "split_time": "3575117,4140678"}, "12": {"slice": true, "split_time": "4140678,4144993"}, "13": {"slice": false, "split_time": "4144993,4648454"}, "14": {"slice": true, "split_time": "4648454,4713755"}, "15": {"slice": false, "split_time": "4713755,5376955"}, "16": {"slice": true, "split_time": "5376955,5439135"}, "17": {"slice": false, "split_time": "5439135,6808121"}, "18": {"slice": true, "split_time": "6808121,7155814"}, "19": {"slice": false, "split_time": "7155814,7431383"}, "20": {"slice": true, "split_time": "7431383,7436381"}, "21": {"slice": false, "split_time": "7436381,7713354"}, "22": {"slice": true, "split_time": "7713354,7722772"}, "23": {"slice": false, "split_time": "7722772,8255613"}, "24": {"slice": true, "split_time": "8255613,8289478"}, "25": {"slice": false, "split_time": "8289478,8807530"}, "26": {"slice": true, "split_time": "8807530,8861466"}, "27": {"slice": false, "split_time": "8861466,11559289"}, "28": {"slice": true, "split_time": "11559289,11579820"}, "29": {"slice": false, "split_time": "11579820,12102866"}, "30": {"slice": true, "split_time": "12102866,12310228"}, "31": {"slice": false, "split_time": "12310228,14242877"}, "32": {"slice": true, "split_time": "14242877,16470528"}}, "time": 1670694818}, "206f3ed94957c275404f384d6b19dde1": {"chunks": {"0": {"slice": true, "split_time": "0,21898"}, "1": {"slice": false, "split_time": "21898,541417"}, "2": {"slice": true, "split_time": "541417,922867"}, "3": {"slice": false, "split_time": "922867,1392503"}, "4": {"slice": true, "split_time": "1392503,1405901"}, "5": {"slice": false, "split_time": "1405901,1627345"}, "6": {"slice": true, "split_time": "1627345,1633183"}, "7": {"slice": false, "split_time": "1633183,1900663"}, "8": {"slice": true, "split_time": "1900663,1906756"}, "9": {"slice": false, "split_time": "1906756,2813575"}, "10": {"slice": true, "split_time": "2813575,2835695"}, "11": {"slice": false, "split_time": "2835695,3620910"}, "12": {"slice": true, "split_time": "3620910,3627896"}, "13": {"slice": false, "split_time": "3627896,4413893"}, "14": {"slice": true, "split_time": "4413893,4420556"}, "15": {"slice": false, "split_time": "4420556,4873361"}, "16": {"slice": true, "split_time": "4873361,4883277"}, "17": {"slice": false, "split_time": "4883277,5143983"}, "18": {"slice": true, "split_time": "5143983,5159518"}, "19": {"slice": false, "split_time": "5159518,6804276"}, "20": {"slice": true, "split_time": "6804276,6810099"}, "21": {"slice": false, "split_time": "6810099,8356887"}, "22": {"slice": true, "split_time": "8356887,8445078"}, "23": {"slice": false, "split_time": "8445078,9662214"}, "24": {"slice": true, "split_time": "9662214,9915170"}, "25": {"slice": false, "split_time": "9915170,10712473"}, "26": {"slice": true, "split_time": "10712473,10976256"}}, "time": 1670696167}, "8a58e2c47ef7998f534d18aedb26d104": {"chunks": {"0": {"slice": true, "split_time": "0,1016974"}, "1": {"slice": false, "split_time": "1016974,4183341"}, "2": {"slice": true, "split_time": "4183341,4195234"}, "3": {"slice": false, "split_time": "4195234,4563111"}, "4": {"slice": true, "split_time": "4563111,4566193"}, "5": {"slice": false, "split_time": "4566193,7734892"}, "6": {"slice": true, "split_time": "7734892,7740172"}, "7": {"slice": false, "split_time": "7740172,8110737"}, "8": {"slice": true, "split_time": "8110737,8120631"}, "9": {"slice": false, "split_time": "8120631,9762874"}, "10": {"slice": true, "split_time": "9762874,9806974"}, "11": {"slice": false, "split_time": "9806974,10232975"}, "12": {"slice": true, "split_time": "10232975,10261053"}, "13": {"slice": false, "split_time": "10261053,10703727"}, "14": {"slice": true, "split_time": "10703727,11484160"}}, "time": 1670698961}, "79fef7af12d80db21f1d449bc8e61b47": {"chunks": {"0": {"slice": true, "split_time": "0,599709"}, "1": {"slice": false, "split_time": "599709,1514834"}, "2": {"slice": true, "split_time": "1514834,1524038"}, "3": {"slice": false, "split_time": "1524038,3814846"}, "4": {"slice": true, "split_time": "3814846,3835226"}, "5": {"slice": false, "split_time": "3835226,7062348"}, "6": {"slice": true, "split_time": "7062348,7129100"}, "7": {"slice": false, "split_time": "7129100,7505629"}, "8": {"slice": true, "split_time": "7505629,7601940"}, "9": {"slice": false, "split_time": "7601940,9801911"}, "10": {"slice": true, "split_time": "9801911,10867200"}}, "time": 1670700629}, "3b55088bb8a3066013ddac86fd55e2ad": {"chunks": {"0": {"slice": true, "split_time": "0,624919"}, "1": {"slice": false, "split_time": "624919,905397"}, "2": {"slice": true, "split_time": "905397,918211"}, "3": {"slice": false, "split_time": "918211,1512000"}, "4": {"slice": true, "split_time": "1512000,1527441"}, "5": {"slice": false, "split_time": "1527441,2424306"}, "6": {"slice": true, "split_time": "2424306,2432801"}, "7": {"slice": false, "split_time": "2432801,3820594"}, "8": {"slice": true, "split_time": "3820594,3840607"}, "9": {"slice": false, "split_time": "3840607,4104884"}, "10": {"slice": true, "split_time": "4104884,4114263"}, "11": {"slice": false, "split_time": "4114263,5027433"}, "12": {"slice": true, "split_time": "5027433,5036666"}, "13": {"slice": false, "split_time": "5036666,7053171"}, "14": {"slice": true, "split_time": "7053171,7604908"}, "15": {"slice": false, "split_time": "7604908,8236806"}, "16": {"slice": true, "split_time": "8236806,8244096"}, "17": {"slice": false, "split_time": "8244096,9793094"}, "18": {"slice": true, "split_time": "9793094,10867200"}}, "time": 1670702000}, "82c947cb69a1be602676351b52d7ee80": {"chunks": {"0": {"slice": true, "split_time": "0,500454"}, "1": {"slice": false, "split_time": "500454,3450969"}, "2": {"slice": true, "split_time": "3450969,3801869"}, "3": {"slice": false, "split_time": "3801869,5043540"}, "4": {"slice": true, "split_time": "5043540,5308663"}, "5": {"slice": false, "split_time": "5308663,6079824"}, "6": {"slice": true, "split_time": "6079824,6256768"}, "7": {"slice": false, "split_time": "6256768,6944231"}, "8": {"slice": true, "split_time": "6944231,6950001"}, "9": {"slice": false, "split_time": "6950001,8637874"}, "10": {"slice": true, "split_time": "8637874,9378240"}}, "time": 1670706107}, "1b8f9bd7b27be8b597ca81e1eea2c42e": {"chunks": {"0": {"slice": true, "split_time": "0,9110"}, "1": {"slice": false, "split_time": "9110,705883"}, "2": {"slice": true, "split_time": "705883,718912"}, "3": {"slice": false, "split_time": "718912,4361016"}, "4": {"slice": true, "split_time": "4361016,4483527"}, "5": {"slice": false, "split_time": "4483527,8236744"}, "6": {"slice": true, "split_time": "8236744,8242271"}, "7": {"slice": false, "split_time": "8242271,8922519"}, "8": {"slice": true, "split_time": "8922519,10047781"}, "9": {"slice": false, "split_time": "10047781,11095794"}, "10": {"slice": true, "split_time": "11095794,11111120"}, "11": {"slice": false, "split_time": "11111120,14067827"}, "12": {"slice": true, "split_time": "14067827,15011040"}}, "time": 1670743675}, "d00da3c641d6ffb90f378fb124158474": {"chunks": {"0": {"slice": true, "split_time": "0,49890"}, "1": {"slice": false, "split_time": "49890,541445"}, "2": {"slice": true, "split_time": "541445,890954"}, "3": {"slice": false, "split_time": "890954,6800207"}, "4": {"slice": true, "split_time": "6800207,6820577"}, "5": {"slice": false, "split_time": "6820577,9448427"}, "6": {"slice": true, "split_time": "9448427,9711360"}}, "time": 1670766134}, "a02e853cfdd66de6264fa2dbdcf7bbb3": {"chunks": {"0": {"slice": true, "split_time": "0,49890"}, "1": {"slice": false, "split_time": "49890,541445"}, "2": {"slice": true, "split_time": "541445,890954"}, "3": {"slice": false, "split_time": "890954,3592747"}}, "time": 1670766345}, "e17e7721e276f0f2198e8b37b5757f8c": {"chunks": {"0": {"slice": true, "split_time": "0,3619457"}, "1": {"slice": false, "split_time": "3619457,6800207"}, "2": {"slice": true, "split_time": "6800207,6820577"}, "3": {"slice": false, "split_time": "6820577,9448427"}, "4": {"slice": true, "split_time": "9448427,9711360"}}, "time": 1670766536}, "24992cb6abdfd24fe115d7b3bdb6a77a": {"chunks": {"0": {"slice": true, "split_time": "0,1140146"}, "1": {"slice": false, "split_time": "1140146,4513827"}, "2": {"slice": true, "split_time": "4513827,4522460"}, "3": {"slice": false, "split_time": "4522460,7854253"}, "4": {"slice": true, "split_time": "7854253,8012303"}, "5": {"slice": false, "split_time": "8012303,8371267"}, "6": {"slice": true, "split_time": "8371267,8674359"}, "7": {"slice": false, "split_time": "8674359,10364809"}, "8": {"slice": true, "split_time": "10364809,12002400"}}, "time": 1670859513}, "d10881af08a0d90a7517585454d22674": {"chunks": {"0": {"slice": true, "split_time": "0,55515"}, "1": {"slice": false, "split_time": "55515,1810232"}, "2": {"slice": true, "split_time": "1810232,1821386"}, "3": {"slice": false, "split_time": "1821386,2687101"}, "4": {"slice": true, "split_time": "2687101,2690771"}, "5": {"slice": false, "split_time": "2690771,5273988"}, "6": {"slice": true, "split_time": "5273988,6165351"}, "7": {"slice": false, "split_time": "6165351,6721480"}, "8": {"slice": true, "split_time": "6721480,6727610"}, "9": {"slice": false, "split_time": "6727610,8962706"}, "10": {"slice": true, "split_time": "8962706,9029280"}}, "time": 1670863486}, "d6aaa656f3dfa79175dd7da1abfaa169": {"chunks": {"0": {"slice": true, "split_time": "0,1058932"}, "1": {"slice": false, "split_time": "1058932,1349051"}, "2": {"slice": true, "split_time": "1349051,1368137"}, "3": {"slice": false, "split_time": "1368137,1662781"}, "4": {"slice": true, "split_time": "1662781,1676436"}, "5": {"slice": false, "split_time": "1676436,2158243"}, "6": {"slice": true, "split_time": "2158243,2162622"}, "7": {"slice": false, "split_time": "2162622,2873063"}, "8": {"slice": true, "split_time": "2873063,2951428"}, "9": {"slice": false, "split_time": "2951428,3262830"}, "10": {"slice": true, "split_time": "3262830,3278072"}, "11": {"slice": false, "split_time": "3278072,3570154"}, "12": {"slice": true, "split_time": "3570154,3591188"}, "13": {"slice": false, "split_time": "3591188,3880090"}, "14": {"slice": true, "split_time": "3880090,3895745"}, "15": {"slice": false, "split_time": "3895745,4368396"}, "16": {"slice": true, "split_time": "4368396,4382892"}, "17": {"slice": false, "split_time": "4382892,4743754"}, "18": {"slice": true, "split_time": "4743754,4747764"}, "19": {"slice": false, "split_time": "4747764,5066193"}, "20": {"slice": true, "split_time": "5066193,5072951"}, "21": {"slice": false, "split_time": "5072951,5240665"}, "22": {"slice": true, "split_time": "5240665,6091680"}}, "time": 1670868921}, "834fb6eb0a9e8af2d8e75a2a97531b34": {"chunks": {"0": {"slice": false, "split_time": "0,144312"}}, "time": 1670939080}, "8170f5afd18e1501b5dbb07ffd008152": {"chunks": {"0": {"slice": true, "split_time": "0,519292"}, "1": {"slice": false, "split_time": "519292,1178914"}, "2": {"slice": true, "split_time": "1178914,1180942"}, "3": {"slice": false, "split_time": "1180942,1861164"}, "4": {"slice": true, "split_time": "1861164,1876075"}, "5": {"slice": false, "split_time": "1876075,2395697"}, "6": {"slice": true, "split_time": "2395697,2401759"}, "7": {"slice": false, "split_time": "2401759,3072500"}, "8": {"slice": true, "split_time": "3072500,3571954"}, "9": {"slice": false, "split_time": "3571954,4107039"}, "10": {"slice": true, "split_time": "4107039,4116965"}, "11": {"slice": false, "split_time": "4116965,4362326"}, "12": {"slice": true, "split_time": "4362326,4367003"}, "13": {"slice": false, "split_time": "4367003,4904071"}, "14": {"slice": true, "split_time": "4904071,4908341"}, "15": {"slice": false, "split_time": "4908341,6337673"}, "16": {"slice": true, "split_time": "6337673,6348520"}, "17": {"slice": false, "split_time": "6348520,6602318"}, "18": {"slice": true, "split_time": "6602318,6609065"}, "19": {"slice": false, "split_time": "6609065,7287032"}, "20": {"slice": true, "split_time": "7287032,7677135"}, "21": {"slice": false, "split_time": "7677135,8138276"}, "22": {"slice": true, "split_time": "8138276,8159058"}, "23": {"slice": false, "split_time": "8159058,8996050"}, "24": {"slice": true, "split_time": "8996050,9751680"}}, "time": 1671033273}, "8ca28e39078a405a79203976e468c50b": {"chunks": {"0": {"slice": true, "split_time": "0,497242"}, "1": {"slice": false, "split_time": "497242,1181163"}, "2": {"slice": true, "split_time": "1181163,1189671"}, "3": {"slice": false, "split_time": "1189671,1849697"}, "4": {"slice": true, "split_time": "1849697,1856537"}, "5": {"slice": false, "split_time": "1856537,3073808"}, "6": {"slice": true, "split_time": "3073808,3557665"}, "7": {"slice": false, "split_time": "3557665,4220637"}, "8": {"slice": true, "split_time": "4220637,4221887"}, "9": {"slice": false, "split_time": "4221887,6334512"}, "10": {"slice": true, "split_time": "6334512,6344041"}, "11": {"slice": false, "split_time": "6344041,6594961"}, "12": {"slice": true, "split_time": "6594961,6598932"}, "13": {"slice": false, "split_time": "6598932,7262331"}, "14": {"slice": true, "split_time": "7262331,7667422"}, "15": {"slice": false, "split_time": "7667422,8133291"}, "16": {"slice": true, "split_time": "8133291,8146462"}, "17": {"slice": false, "split_time": "8146462,8995577"}, "18": {"slice": true, "split_time": "8995577,9748320"}}, "time": 1671036817}, "37add5fae144aabb15b2698a775acb9f": {"chunks": {"0": {"slice": true, "split_time": "0,552464"}, "1": {"slice": false, "split_time": "552464,1664737"}, "2": {"slice": true, "split_time": "1664737,1713934"}, "3": {"slice": false, "split_time": "1713934,2128629"}, "4": {"slice": true, "split_time": "2128629,2190888"}, "5": {"slice": false, "split_time": "2190888,2467571"}, "6": {"slice": true, "split_time": "2467571,2492274"}, "7": {"slice": false, "split_time": "2492274,2786017"}, "8": {"slice": true, "split_time": "2786017,2888014"}, "9": {"slice": false, "split_time": "2888014,3126127"}, "10": {"slice": true, "split_time": "3126127,3192892"}, "11": {"slice": false, "split_time": "3192892,3462081"}, "12": {"slice": true, "split_time": "3462081,3510231"}, "13": {"slice": false, "split_time": "3510231,4088592"}, "14": {"slice": true, "split_time": "4088592,4131082"}, "15": {"slice": false, "split_time": "4131082,5405140"}, "16": {"slice": true, "split_time": "5405140,5444945"}, "17": {"slice": false, "split_time": "5444945,5744967"}, "18": {"slice": true, "split_time": "5744967,5958048"}, "19": {"slice": false, "split_time": "5958048,6530239"}, "20": {"slice": true, "split_time": "6530239,6583390"}, "21": {"slice": false, "split_time": "6583390,6858361"}, "22": {"slice": true, "split_time": "6858361,6902846"}, "23": {"slice": false, "split_time": "6902846,8834591"}, "24": {"slice": true, "split_time": "8834591,8837998"}, "25": {"slice": false, "split_time": "8837998,9132266"}, "26": {"slice": true, "split_time": "9132266,9668615"}, "27": {"slice": false, "split_time": "9668615,10936654"}, "28": {"slice": true, "split_time": "10936654,10951534"}, "29": {"slice": false, "split_time": "10951534,11236384"}, "30": {"slice": true, "split_time": "11236384,11381600"}, "31": {"slice": false, "split_time": "11381600,11577220"}, "32": {"slice": true, "split_time": "11577220,12037440"}}, "time": 1671109941}, "93d732f016d67271349fba5d1e75c18d": {"chunks": {"0": {"slice": true, "split_time": "0,8804"}, "1": {"slice": false, "split_time": "8804,438688"}, "2": {"slice": true, "split_time": "438688,486543"}, "3": {"slice": false, "split_time": "486543,1870403"}, "4": {"slice": true, "split_time": "1870403,1877776"}, "5": {"slice": false, "split_time": "1877776,3930315"}, "6": {"slice": true, "split_time": "3930315,4063338"}, "7": {"slice": false, "split_time": "4063338,4394494"}, "8": {"slice": true, "split_time": "4394494,4399789"}, "9": {"slice": false, "split_time": "4399789,5046717"}, "10": {"slice": true, "split_time": "5046717,5059510"}, "11": {"slice": false, "split_time": "5059510,5621793"}, "12": {"slice": true, "split_time": "5621793,5633688"}, "13": {"slice": false, "split_time": "5633688,6917893"}, "14": {"slice": true, "split_time": "6917893,6923244"}, "15": {"slice": false, "split_time": "6923244,7553373"}, "16": {"slice": true, "split_time": "7553373,7689423"}, "17": {"slice": false, "split_time": "7689423,8079372"}, "18": {"slice": true, "split_time": "8079372,8080948"}, "19": {"slice": false, "split_time": "8080948,8344057"}, "20": {"slice": true, "split_time": "8344057,8383507"}, "21": {"slice": false, "split_time": "8383507,8862061"}, "22": {"slice": true, "split_time": "8862061,8865820"}, "23": {"slice": false, "split_time": "8865820,9257710"}, "24": {"slice": true, "split_time": "9257710,9370835"}, "25": {"slice": false, "split_time": "9370835,10125170"}, "26": {"slice": true, "split_time": "10125170,10142843"}, "27": {"slice": false, "split_time": "10142843,11858753"}, "28": {"slice": true, "split_time": "11858753,12052800"}}, "time": 1671113987}, "751d0ea9f160787c40c810f367290353": {"chunks": {"0": {"slice": true, "split_time": "0,1579118"}, "1": {"slice": false, "split_time": "1579118,1820807"}, "2": {"slice": true, "split_time": "1820807,1873005"}, "3": {"slice": false, "split_time": "1873005,2144972"}, "4": {"slice": true, "split_time": "2144972,2176539"}, "5": {"slice": false, "split_time": "2176539,2486160"}, "6": {"slice": true, "split_time": "2486160,2489237"}, "7": {"slice": false, "split_time": "2489237,4090074"}, "8": {"slice": true, "split_time": "4090074,4450548"}, "9": {"slice": false, "split_time": "4450548,4714769"}, "10": {"slice": true, "split_time": "4714769,4751693"}, "11": {"slice": false, "split_time": "4751693,5019002"}, "12": {"slice": true, "split_time": "5019002,5050186"}, "13": {"slice": false, "split_time": "5050186,5356688"}, "14": {"slice": true, "split_time": "5356688,5365172"}, "15": {"slice": false, "split_time": "5365172,6967578"}, "16": {"slice": true, "split_time": "6967578,7620170"}, "17": {"slice": false, "split_time": "7620170,8820345"}, "18": {"slice": true, "split_time": "8820345,8834237"}, "19": {"slice": false, "split_time": "8834237,10763919"}, "20": {"slice": true, "split_time": "10763919,10780703"}, "21": {"slice": false, "split_time": "10780703,10930737"}, "22": {"slice": true, "split_time": "10930737,11670240"}}, "time": 1671122543}, "01160d6a8923cb22b9fcb057152d77c4": {"chunks": {"0": {"slice": true, "split_time": "0,1570889"}, "1": {"slice": false, "split_time": "1570889,1818327"}, "2": {"slice": true, "split_time": "1818327,1865958"}, "3": {"slice": false, "split_time": "1865958,2142600"}, "4": {"slice": true, "split_time": "2142600,2179347"}, "5": {"slice": false, "split_time": "2179347,2482440"}, "6": {"slice": true, "split_time": "2482440,2489004"}, "7": {"slice": false, "split_time": "2489004,4083686"}, "8": {"slice": true, "split_time": "4083686,4448991"}, "9": {"slice": false, "split_time": "4448991,4713912"}, "10": {"slice": true, "split_time": "4713912,4755720"}, "11": {"slice": false, "split_time": "4755720,5015037"}, "12": {"slice": true, "split_time": "5015037,5046087"}, "13": {"slice": false, "split_time": "5046087,5357051"}, "14": {"slice": true, "split_time": "5357051,5362838"}, "15": {"slice": false, "split_time": "5362838,6971462"}, "16": {"slice": true, "split_time": "6971462,7615732"}, "17": {"slice": false, "split_time": "7615732,10929758"}, "18": {"slice": true, "split_time": "10929758,11664480"}}, "time": 1671124339}, "5cbf5f07cf04349c7c9b184e3f964024": {"chunks": {"0": {"slice": true, "split_time": "0,691826"}, "1": {"slice": false, "split_time": "691826,1876389"}, "2": {"slice": true, "split_time": "1876389,1911382"}, "3": {"slice": false, "split_time": "1911382,4282343"}, "4": {"slice": true, "split_time": "4282343,4517196"}, "5": {"slice": false, "split_time": "4517196,5711146"}, "6": {"slice": true, "split_time": "5711146,5741551"}, "7": {"slice": false, "split_time": "5741551,8086325"}, "8": {"slice": true, "split_time": "8086325,9003110"}, "9": {"slice": false, "split_time": "9003110,11470072"}, "10": {"slice": true, "split_time": "11470072,13332960"}}, "time": 1671238083}, "e64b595ad8810c907b25d12bd57c3904": {"chunks": {"0": {"slice": true, "split_time": "0,36395"}, "1": {"slice": false, "split_time": "36395,1392029"}, "2": {"slice": true, "split_time": "1392029,1936181"}, "3": {"slice": false, "split_time": "1936181,3142145"}, "4": {"slice": true, "split_time": "3142145,3167534"}, "5": {"slice": false, "split_time": "3167534,5527939"}, "6": {"slice": true, "split_time": "5527939,5576252"}, "7": {"slice": false, "split_time": "5576252,6960530"}, "8": {"slice": true, "split_time": "6960530,6988768"}, "9": {"slice": false, "split_time": "6988768,9356930"}, "10": {"slice": true, "split_time": "9356930,9800850"}, "11": {"slice": false, "split_time": "9800850,12723584"}, "12": {"slice": true, "split_time": "12723584,12892059"}, "13": {"slice": false, "split_time": "12892059,13545240"}, "14": {"slice": true, "split_time": "13545240,13627603"}, "15": {"slice": false, "split_time": "13627603,14337070"}, "16": {"slice": true, "split_time": "14337070,14595072"}}, "time": 1671245700}}

inference/infer_tool.py

@@ -0,0 +1,327 @@
+import hashlib
+import json
+import logging
+import os
+import time
+from pathlib import Path
+
+import librosa
+import maad
+import numpy as np
+# import onnxruntime
+import parselmouth
+import soundfile
+import torch
+import torchaudio
+
+from hubert import hubert_model
+import utils
+from models import SynthesizerTrn
+
+logging.getLogger('matplotlib').setLevel(logging.WARNING)
+
+
+def read_temp(file_name):
+    if not os.path.exists(file_name):
+        with open(file_name, "w") as f:
+            f.write(json.dumps({"info": "temp_dict"}))
+        return {}
+    else:
+        try:
+            with open(file_name, "r") as f:
+                data = f.read()
+            data_dict = json.loads(data)
+            if os.path.getsize(file_name) > 50 * 1024 * 1024:
+                f_name = file_name.split("/")[-1]
+                print(f"clean {f_name}")
+                for wav_hash in list(data_dict.keys()):
+                    if int(time.time()) - int(data_dict[wav_hash]["time"]) > 14 * 24 * 3600:
+                        del data_dict[wav_hash]
+        except Exception as e:
+            print(e)
+            print(f"{file_name} error,auto rebuild file")
+            data_dict = {"info": "temp_dict"}
+        return data_dict
+
+
+def write_temp(file_name, data):
+    with open(file_name, "w") as f:
+        f.write(json.dumps(data))
+
+
+def timeit(func):
+    def run(*args, **kwargs):
+        t = time.time()
+        res = func(*args, **kwargs)
+        print('executing \'%s\' costed %.3fs' % (func.__name__, time.time() - t))
+        return res
+
+    return run
+
+
+def format_wav(audio_path):
+    if Path(audio_path).suffix == '.wav':
+        return
+    raw_audio, raw_sample_rate = librosa.load(audio_path, mono=True, sr=None)
+    soundfile.write(Path(audio_path).with_suffix(".wav"), raw_audio, raw_sample_rate)
+
+
+def get_end_file(dir_path, end):
+    file_lists = []
+    for root, dirs, files in os.walk(dir_path):
+        files = [f for f in files if f[0] != '.']
+        dirs[:] = [d for d in dirs if d[0] != '.']
+        for f_file in files:
+            if f_file.endswith(end):
+                file_lists.append(os.path.join(root, f_file).replace("\\", "/"))
+    return file_lists
+
+
+def get_md5(content):
+    return hashlib.new("md5", content).hexdigest()
+
+
+def resize2d_f0(x, target_len):
+    source = np.array(x)
+    source[source < 0.001] = np.nan
+    target = np.interp(np.arange(0, len(source) * target_len, len(source)) / target_len, np.arange(0, len(source)),
+                       source)
+    res = np.nan_to_num(target)
+    return res
+
+def get_f0(x, p_len,f0_up_key=0):
+
+    time_step = 160 / 16000 * 1000
+    f0_min = 50
+    f0_max = 1100
+    f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+    f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+
+    f0 = parselmouth.Sound(x, 16000).to_pitch_ac(
+        time_step=time_step / 1000, voicing_threshold=0.6,
+        pitch_floor=f0_min, pitch_ceiling=f0_max).selected_array['frequency']
+
+    pad_size=(p_len - len(f0) + 1) // 2
+    if(pad_size>0 or p_len - len(f0) - pad_size>0):
+        f0 = np.pad(f0,[[pad_size,p_len - len(f0) - pad_size]], mode='constant')
+
+    f0 *= pow(2, f0_up_key / 12)
+    f0_mel = 1127 * np.log(1 + f0 / 700)
+    f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (f0_mel_max - f0_mel_min) + 1
+    f0_mel[f0_mel <= 1] = 1
+    f0_mel[f0_mel > 255] = 255
+    f0_coarse = np.rint(f0_mel).astype(np.int)
+    return f0_coarse, f0
+
+def clean_pitch(input_pitch):
+    num_nan = np.sum(input_pitch == 1)
+    if num_nan / len(input_pitch) > 0.9:
+        input_pitch[input_pitch != 1] = 1
+    return input_pitch
+
+
+def plt_pitch(input_pitch):
+    input_pitch = input_pitch.astype(float)
+    input_pitch[input_pitch == 1] = np.nan
+    return input_pitch
+
+
+def f0_to_pitch(ff):
+    f0_pitch = 69 + 12 * np.log2(ff / 440)
+    return f0_pitch
+
+
+def fill_a_to_b(a, b):
+    if len(a) < len(b):
+        for _ in range(0, len(b) - len(a)):
+            a.append(a[0])
+
+
+def mkdir(paths: list):
+    for path in paths:
+        if not os.path.exists(path):
+            os.mkdir(path)
+
+
+class Svc(object):
+    def __init__(self, net_g_path, config_path, hubert_path="hubert/hubert-soft-0d54a1f4.pt",
+                 onnx=False):
+        self.onnx = onnx
+        self.net_g_path = net_g_path
+        self.hubert_path = hubert_path
+        self.dev = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.net_g_ms = None
+        self.hps_ms = utils.get_hparams_from_file(config_path)
+        self.target_sample = self.hps_ms.data.sampling_rate
+        self.hop_size = self.hps_ms.data.hop_length
+        self.speakers = {}
+        for spk, sid in self.hps_ms.spk.items():
+            self.speakers[sid] = spk
+        self.spk2id = self.hps_ms.spk
+        # 加载hubert
+        self.hubert_soft = hubert_model.hubert_soft(hubert_path)
+        if torch.cuda.is_available():
+            self.hubert_soft = self.hubert_soft.cuda()
+        self.load_model()
+
+    def load_model(self):
+        # 获取模型配置
+        if self.onnx:
+            raise NotImplementedError
+            # self.net_g_ms = SynthesizerTrnForONNX(
+            #     178,
+            #     self.hps_ms.data.filter_length // 2 + 1,
+            #     self.hps_ms.train.segment_size // self.hps_ms.data.hop_length,
+            #     n_speakers=self.hps_ms.data.n_speakers,
+            #     **self.hps_ms.model)
+            # _ = utils.load_checkpoint(self.net_g_path, self.net_g_ms, None)
+        else:
+            self.net_g_ms = SynthesizerTrn(
+                self.hps_ms.data.filter_length // 2 + 1,
+                self.hps_ms.train.segment_size // self.hps_ms.data.hop_length,
+                **self.hps_ms.model)
+            _ = utils.load_checkpoint(self.net_g_path, self.net_g_ms, None)
+        if "half" in self.net_g_path and torch.cuda.is_available():
+            _ = self.net_g_ms.half().eval().to(self.dev)
+        else:
+            _ = self.net_g_ms.eval().to(self.dev)
+
+    def get_units(self, source, sr):
+
+        source = source.unsqueeze(0).to(self.dev)
+        with torch.inference_mode():
+            start = time.time()
+            units = self.hubert_soft.units(source)
+            use_time = time.time() - start
+            print("hubert use time:{}".format(use_time))
+            return units
+
+
+    def get_unit_pitch(self, in_path, tran):
+        source, sr = torchaudio.load(in_path)
+        source = torchaudio.functional.resample(source, sr, 16000)
+        if len(source.shape) == 2 and source.shape[1] >= 2:
+            source = torch.mean(source, dim=0).unsqueeze(0)
+        soft = self.get_units(source, sr).squeeze(0).cpu().numpy()
+        f0_coarse, f0 = get_f0(source.cpu().numpy()[0], soft.shape[0]*2, tran)
+        f0 = resize2d_f0(f0, soft.shape[0]*3)
+        return soft, f0
+
+    def infer(self, speaker_id, tran, raw_path):
+        if type(speaker_id) == str:
+            speaker_id = self.spk2id[speaker_id]
+        sid = torch.LongTensor([int(speaker_id)]).to(self.dev).unsqueeze(0)
+        soft, pitch = self.get_unit_pitch(raw_path, tran)
+        f0 = torch.FloatTensor(clean_pitch(pitch)).unsqueeze(0).to(self.dev)
+        if "half" in self.net_g_path and torch.cuda.is_available():
+            stn_tst = torch.HalfTensor(soft)
+        else:
+            stn_tst = torch.FloatTensor(soft)
+        with torch.no_grad():
+            x_tst = stn_tst.unsqueeze(0).to(self.dev)
+            start = time.time()
+            x_tst = torch.repeat_interleave(x_tst, repeats=3, dim=1).transpose(1, 2)
+            audio = self.net_g_ms.infer(x_tst, f0=f0, g=sid)[0,0].data.float()
+            use_time = time.time() - start
+            print("vits use time:{}".format(use_time))
+        return audio, audio.shape[-1]
+
+
+# class SvcONNXInferModel(object):
+#     def __init__(self, hubert_onnx, vits_onnx, config_path):
+#         self.config_path = config_path
+#         self.vits_onnx = vits_onnx
+#         self.hubert_onnx = hubert_onnx
+#         self.hubert_onnx_session = onnxruntime.InferenceSession(hubert_onnx, providers=['CUDAExecutionProvider', ])
+#         self.inspect_onnx(self.hubert_onnx_session)
+#         self.vits_onnx_session = onnxruntime.InferenceSession(vits_onnx, providers=['CUDAExecutionProvider', ])
+#         self.inspect_onnx(self.vits_onnx_session)
+#         self.hps_ms = utils.get_hparams_from_file(self.config_path)
+#         self.target_sample = self.hps_ms.data.sampling_rate
+#         self.feature_input = FeatureInput(self.hps_ms.data.sampling_rate, self.hps_ms.data.hop_length)
+#
+#     @staticmethod
+#     def inspect_onnx(session):
+#         for i in session.get_inputs():
+#             print("name:{}\tshape:{}\tdtype:{}".format(i.name, i.shape, i.type))
+#         for i in session.get_outputs():
+#             print("name:{}\tshape:{}\tdtype:{}".format(i.name, i.shape, i.type))
+#
+#     def infer(self, speaker_id, tran, raw_path):
+#         sid = np.array([int(speaker_id)], dtype=np.int64)
+#         soft, pitch = self.get_unit_pitch(raw_path, tran)
+#         pitch = np.expand_dims(pitch, axis=0).astype(np.int64)
+#         stn_tst = soft
+#         x_tst = np.expand_dims(stn_tst, axis=0)
+#         x_tst_lengths = np.array([stn_tst.shape[0]], dtype=np.int64)
+#         # 使用ONNX Runtime进行推理
+#         start = time.time()
+#         audio = self.vits_onnx_session.run(output_names=["audio"],
+#                                            input_feed={
+#                                                "hidden_unit": x_tst,
+#                                                "lengths": x_tst_lengths,
+#                                                "pitch": pitch,
+#                                                "sid": sid,
+#                                            })[0][0, 0]
+#         use_time = time.time() - start
+#         print("vits_onnx_session.run time:{}".format(use_time))
+#         audio = torch.from_numpy(audio)
+#         return audio, audio.shape[-1]
+#
+#     def get_units(self, source, sr):
+#         source = torchaudio.functional.resample(source, sr, 16000)
+#         if len(source.shape) == 2 and source.shape[1] >= 2:
+#             source = torch.mean(source, dim=0).unsqueeze(0)
+#         source = source.unsqueeze(0)
+#         # 使用ONNX Runtime进行推理
+#         start = time.time()
+#         units = self.hubert_onnx_session.run(output_names=["embed"],
+#                                              input_feed={"source": source.numpy()})[0]
+#         use_time = time.time() - start
+#         print("hubert_onnx_session.run time:{}".format(use_time))
+#         return units
+#
+#     def transcribe(self, source, sr, length, transform):
+#         feature_pit = self.feature_input.compute_f0(source, sr)
+#         feature_pit = feature_pit * 2 ** (transform / 12)
+#         feature_pit = resize2d_f0(feature_pit, length)
+#         coarse_pit = self.feature_input.coarse_f0(feature_pit)
+#         return coarse_pit
+#
+#     def get_unit_pitch(self, in_path, tran):
+#         source, sr = torchaudio.load(in_path)
+#         soft = self.get_units(source, sr).squeeze(0)
+#         input_pitch = self.transcribe(source.numpy()[0], sr, soft.shape[0], tran)
+#         return soft, input_pitch
+
+
+class RealTimeVC:
+    def __init__(self):
+        self.last_chunk = None
+        self.last_o = None
+        self.chunk_len = 16000  # 区块长度
+        self.pre_len = 3840  # 交叉淡化长度，640的倍数
+
+    """输入输出都是1维numpy 音频波形数组"""
+
+    def process(self, svc_model, speaker_id, f_pitch_change, input_wav_path):
+        audio, sr = torchaudio.load(input_wav_path)
+        audio = audio.cpu().numpy()[0]
+        temp_wav = io.BytesIO()
+        if self.last_chunk is None:
+            input_wav_path.seek(0)
+            audio, sr = svc_model.infer(speaker_id, f_pitch_change, input_wav_path)
+            audio = audio.cpu().numpy()
+            self.last_chunk = audio[-self.pre_len:]
+            self.last_o = audio
+            return audio[-self.chunk_len:]
+        else:
+            audio = np.concatenate([self.last_chunk, audio])
+            soundfile.write(temp_wav, audio, sr, format="wav")
+            temp_wav.seek(0)
+            audio, sr = svc_model.infer(speaker_id, f_pitch_change, temp_wav)
+            audio = audio.cpu().numpy()
+            ret = maad.util.crossfade(self.last_o, audio, self.pre_len)
+            self.last_chunk = audio[-self.pre_len:]
+            self.last_o = audio
+            return ret[self.chunk_len:2 * self.chunk_len]

inference/slicer.py

@@ -0,0 +1,158 @@
+import time
+
+import numpy as np
+import torch
+import torchaudio
+from scipy.ndimage import maximum_filter1d, uniform_filter1d
+
+
+def timeit(func):
+    def run(*args, **kwargs):
+        t = time.time()
+        res = func(*args, **kwargs)
+        print('executing \'%s\' costed %.3fs' % (func.__name__, time.time() - t))
+        return res
+
+    return run
+
+
+# @timeit
+def _window_maximum(arr, win_sz):
+    return maximum_filter1d(arr, size=win_sz)[win_sz // 2: win_sz // 2 + arr.shape[0] - win_sz + 1]
+
+
+# @timeit
+def _window_rms(arr, win_sz):
+    filtered = np.sqrt(uniform_filter1d(np.power(arr, 2), win_sz) - np.power(uniform_filter1d(arr, win_sz), 2))
+    return filtered[win_sz // 2: win_sz // 2 + arr.shape[0] - win_sz + 1]
+
+
+def level2db(levels, eps=1e-12):
+    return 20 * np.log10(np.clip(levels, a_min=eps, a_max=1))
+
+
+def _apply_slice(audio, begin, end):
+    if len(audio.shape) > 1:
+        return audio[:, begin: end]
+    else:
+        return audio[begin: end]
+
+
+class Slicer:
+    def __init__(self,
+                 sr: int,
+                 db_threshold: float = -40,
+                 min_length: int = 5000,
+                 win_l: int = 300,
+                 win_s: int = 20,
+                 max_silence_kept: int = 500):
+        self.db_threshold = db_threshold
+        self.min_samples = round(sr * min_length / 1000)
+        self.win_ln = round(sr * win_l / 1000)
+        self.win_sn = round(sr * win_s / 1000)
+        self.max_silence = round(sr * max_silence_kept / 1000)
+        if not self.min_samples >= self.win_ln >= self.win_sn:
+            raise ValueError('The following condition must be satisfied: min_length >= win_l >= win_s')
+        if not self.max_silence >= self.win_sn:
+            raise ValueError('The following condition must be satisfied: max_silence_kept >= win_s')
+
+    @timeit
+    def slice(self, audio):
+        samples = audio
+        if samples.shape[0] <= self.min_samples:
+            return {"0": {"slice": False, "split_time": f"0,{len(audio)}"}}
+        # get absolute amplitudes
+        abs_amp = np.abs(samples - np.mean(samples))
+        # calculate local maximum with large window
+        win_max_db = level2db(_window_maximum(abs_amp, win_sz=self.win_ln))
+        sil_tags = []
+        left = right = 0
+        while right < win_max_db.shape[0]:
+            if win_max_db[right] < self.db_threshold:
+                right += 1
+            elif left == right:
+                left += 1
+                right += 1
+            else:
+                if left == 0:
+                    split_loc_l = left
+                else:
+                    sil_left_n = min(self.max_silence, (right + self.win_ln - left) // 2)
+                    rms_db_left = level2db(_window_rms(samples[left: left + sil_left_n], win_sz=self.win_sn))
+                    split_win_l = left + np.argmin(rms_db_left)
+                    split_loc_l = split_win_l + np.argmin(abs_amp[split_win_l: split_win_l + self.win_sn])
+                if len(sil_tags) != 0 and split_loc_l - sil_tags[-1][1] < self.min_samples and right < win_max_db.shape[
+                    0] - 1:
+                    right += 1
+                    left = right
+                    continue
+                if right == win_max_db.shape[0] - 1:
+                    split_loc_r = right + self.win_ln
+                else:
+                    sil_right_n = min(self.max_silence, (right + self.win_ln - left) // 2)
+                    rms_db_right = level2db(_window_rms(samples[right + self.win_ln - sil_right_n: right + self.win_ln],
+                                                        win_sz=self.win_sn))
+                    split_win_r = right + self.win_ln - sil_right_n + np.argmin(rms_db_right)
+                    split_loc_r = split_win_r + np.argmin(abs_amp[split_win_r: split_win_r + self.win_sn])
+                sil_tags.append((split_loc_l, split_loc_r))
+                right += 1
+                left = right
+        if left != right:
+            sil_left_n = min(self.max_silence, (right + self.win_ln - left) // 2)
+            rms_db_left = level2db(_window_rms(samples[left: left + sil_left_n], win_sz=self.win_sn))
+            split_win_l = left + np.argmin(rms_db_left)
+            split_loc_l = split_win_l + np.argmin(abs_amp[split_win_l: split_win_l + self.win_sn])
+            sil_tags.append((split_loc_l, samples.shape[0]))
+        if len(sil_tags) == 0:
+            return {"0": {"slice": False, "split_time": f"0,{len(audio)}"}}
+        else:
+            chunks = []
+            # 第一段静音并非从头开始，补上有声片段
+            if sil_tags[0][0]:
+                chunks.append({"slice": False, "split_time": f"0,{sil_tags[0][0]}"})
+            for i in range(0, len(sil_tags)):
+                # 标识有声片段（跳过第一段）
+                if i:
+                    chunks.append({"slice": False, "split_time": f"{sil_tags[i - 1][1]},{sil_tags[i][0]}"})
+                # 标识所有静音片段
+                chunks.append({"slice": True, "split_time": f"{sil_tags[i][0]},{sil_tags[i][1]}"})
+            # 最后一段静音并非结尾，补上结尾片段
+            if sil_tags[-1][1] != len(audio):
+                chunks.append({"slice": False, "split_time": f"{sil_tags[-1][1]},{len(audio)}"})
+            chunk_dict = {}
+            for i in range(len(chunks)):
+                chunk_dict[str(i)] = chunks[i]
+            return chunk_dict
+
+
+def cut(audio_path, db_thresh=-30, min_len=5000, win_l=300, win_s=20, max_sil_kept=500):
+    audio, sr = torchaudio.load(audio_path)
+    if len(audio.shape) == 2 and audio.shape[1] >= 2:
+        audio = torch.mean(audio, dim=0).unsqueeze(0)
+    audio = audio.cpu().numpy()[0]
+
+    slicer = Slicer(
+        sr=sr,
+        db_threshold=db_thresh,
+        min_length=min_len,
+        win_l=win_l,
+        win_s=win_s,
+        max_silence_kept=max_sil_kept
+    )
+    chunks = slicer.slice(audio)
+    return chunks
+
+
+def chunks2audio(audio_path, chunks):
+    chunks = dict(chunks)
+    audio, sr = torchaudio.load(audio_path)
+    if len(audio.shape) == 2 and audio.shape[1] >= 2:
+        audio = torch.mean(audio, dim=0).unsqueeze(0)
+    audio = audio.cpu().numpy()[0]
+    result = []
+    for k, v in chunks.items():
+        tag = v["split_time"].split(",")
+        result.append((v["slice"], audio[int(tag[0]):int(tag[1])]))
+    return result, sr
+
+

logs/48k/aqua.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6045347e929c571751fa53e789b277f7737c2d53dae53e753b918af6503d2155
+size 699505437

losses.py

@@ -0,0 +1,61 @@
+import torch 
+from torch.nn import functional as F
+
+import commons
+
+
+def feature_loss(fmap_r, fmap_g):
+  loss = 0
+  for dr, dg in zip(fmap_r, fmap_g):
+    for rl, gl in zip(dr, dg):
+      rl = rl.float().detach()
+      gl = gl.float()
+      loss += torch.mean(torch.abs(rl - gl))
+
+  return loss * 2 
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+  loss = 0
+  r_losses = []
+  g_losses = []
+  for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+    dr = dr.float()
+    dg = dg.float()
+    r_loss = torch.mean((1-dr)**2)
+    g_loss = torch.mean(dg**2)
+    loss += (r_loss + g_loss)
+    r_losses.append(r_loss.item())
+    g_losses.append(g_loss.item())
+
+  return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+  loss = 0
+  gen_losses = []
+  for dg in disc_outputs:
+    dg = dg.float()
+    l = torch.mean((1-dg)**2)
+    gen_losses.append(l)
+    loss += l
+
+  return loss, gen_losses
+
+
+def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
+  """
+  z_p, logs_q: [b, h, t_t]
+  m_p, logs_p: [b, h, t_t]
+  """
+  z_p = z_p.float()
+  logs_q = logs_q.float()
+  m_p = m_p.float()
+  logs_p = logs_p.float()
+  z_mask = z_mask.float()
+  #print(logs_p)
+  kl = logs_p - logs_q - 0.5
+  kl += 0.5 * ((z_p - m_p)**2) * torch.exp(-2. * logs_p)
+  kl = torch.sum(kl * z_mask)
+  l = kl / torch.sum(z_mask)
+  return l

mel_processing.py

@@ -0,0 +1,112 @@
+import math
+import os
+import random
+import torch
+from torch import nn
+import torch.nn.functional as F
+import torch.utils.data
+import numpy as np
+import librosa
+import librosa.util as librosa_util
+from librosa.util import normalize, pad_center, tiny
+from scipy.signal import get_window
+from scipy.io.wavfile import read
+from librosa.filters import mel as librosa_mel_fn
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    output = dynamic_range_compression_torch(magnitudes)
+    return output
+
+
+def spectral_de_normalize_torch(magnitudes):
+    output = dynamic_range_decompression_torch(magnitudes)
+    return output
+
+
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    global mel_basis
+    dtype_device = str(spec.dtype) + '_' + str(spec.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(dtype=spec.dtype, device=spec.device)
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+    return spec
+
+
+def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False):
+    if torch.min(y) < -1.:
+        print('min value is ', torch.min(y))
+    if torch.max(y) > 1.:
+        print('max value is ', torch.max(y))
+
+    global mel_basis, hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    fmax_dtype_device = str(fmax) + '_' + dtype_device
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(dtype=y.dtype, device=y.device)
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = y.squeeze(1)
+
+    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+
+    return spec

models.py

@@ -0,0 +1,351 @@
+import copy
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import attentions
+import commons
+import modules
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from commons import init_weights, get_padding
+from vdecoder.hifigan.models import Generator
+from utils import f0_to_coarse
+
+class ResidualCouplingBlock(nn.Module):
+  def __init__(self,
+      channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      n_flows=4,
+      gin_channels=0):
+    super().__init__()
+    self.channels = channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.n_flows = n_flows
+    self.gin_channels = gin_channels
+
+    self.flows = nn.ModuleList()
+    for i in range(n_flows):
+      self.flows.append(modules.ResidualCouplingLayer(channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True))
+      self.flows.append(modules.Flip())
+
+  def forward(self, x, x_mask, g=None, reverse=False):
+    if not reverse:
+      for flow in self.flows:
+        x, _ = flow(x, x_mask, g=g, reverse=reverse)
+    else:
+      for flow in reversed(self.flows):
+        x = flow(x, x_mask, g=g, reverse=reverse)
+    return x
+
+
+class Encoder(nn.Module):
+  def __init__(self,
+      in_channels,
+      out_channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      gin_channels=0):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.gin_channels = gin_channels
+
+    self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+    self.enc = modules.WN(hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels)
+    self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+  def forward(self, x, x_lengths, g=None):
+    # print(x.shape,x_lengths.shape)
+    x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
+    x = self.pre(x) * x_mask
+    x = self.enc(x, x_mask, g=g)
+    stats = self.proj(x) * x_mask
+    m, logs = torch.split(stats, self.out_channels, dim=1)
+    z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+    return z, m, logs, x_mask
+
+
+class TextEncoder(nn.Module):
+  def __init__(self,
+      in_channels,
+      out_channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      gin_channels=0,
+      filter_channels=None,
+      n_heads=None,
+      p_dropout=None):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.gin_channels = gin_channels
+    self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+    self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+    self.f0_emb = nn.Embedding(256, hidden_channels)
+
+    self.enc_ =  attentions.Encoder(
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout)
+
+  def forward(self, x, x_lengths, f0=None):
+    x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
+    x = self.pre(x) * x_mask
+    x = x + self.f0_emb(f0).transpose(1,2)
+    x = self.enc_(x * x_mask, x_mask)
+    stats = self.proj(x) * x_mask
+    m, logs = torch.split(stats, self.out_channels, dim=1)
+    z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+
+    return z, m, logs, x_mask
+
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
+            norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(get_padding(kernel_size, 1), 0))),
+        ])
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0: # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+            norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+            norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+            norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+            norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+        ])
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2,3,5,7,11]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+        
+        
+class SpeakerEncoder(torch.nn.Module):
+    def __init__(self, mel_n_channels=80, model_num_layers=3, model_hidden_size=256, model_embedding_size=256):
+        super(SpeakerEncoder, self).__init__()
+        self.lstm = nn.LSTM(mel_n_channels, model_hidden_size, model_num_layers, batch_first=True)
+        self.linear = nn.Linear(model_hidden_size, model_embedding_size)
+        self.relu = nn.ReLU()
+
+    def forward(self, mels):
+        self.lstm.flatten_parameters()
+        _, (hidden, _) = self.lstm(mels)
+        embeds_raw = self.relu(self.linear(hidden[-1]))
+        return embeds_raw / torch.norm(embeds_raw, dim=1, keepdim=True)
+        
+    def compute_partial_slices(self, total_frames, partial_frames, partial_hop):
+        mel_slices = []
+        for i in range(0, total_frames-partial_frames, partial_hop):
+            mel_range = torch.arange(i, i+partial_frames)
+            mel_slices.append(mel_range)
+            
+        return mel_slices
+    
+    def embed_utterance(self, mel, partial_frames=128, partial_hop=64):
+        mel_len = mel.size(1)
+        last_mel = mel[:,-partial_frames:]
+        
+        if mel_len > partial_frames:
+            mel_slices = self.compute_partial_slices(mel_len, partial_frames, partial_hop)
+            mels = list(mel[:,s] for s in mel_slices)
+            mels.append(last_mel)
+            mels = torch.stack(tuple(mels), 0).squeeze(1)
+        
+            with torch.no_grad():
+                partial_embeds = self(mels)
+            embed = torch.mean(partial_embeds, axis=0).unsqueeze(0)
+            #embed = embed / torch.linalg.norm(embed, 2)
+        else:
+            with torch.no_grad():
+                embed = self(last_mel)
+        
+        return embed
+
+
+class SynthesizerTrn(nn.Module):
+  """
+  Synthesizer for Training
+  """
+
+  def __init__(self, 
+    spec_channels,
+    segment_size,
+    inter_channels,
+    hidden_channels,
+    filter_channels,
+    n_heads,
+    n_layers,
+    kernel_size,
+    p_dropout,
+    resblock, 
+    resblock_kernel_sizes, 
+    resblock_dilation_sizes, 
+    upsample_rates, 
+    upsample_initial_channel, 
+    upsample_kernel_sizes,
+    gin_channels,
+    ssl_dim,
+    n_speakers,
+    **kwargs):
+
+    super().__init__()
+    self.spec_channels = spec_channels
+    self.inter_channels = inter_channels
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.resblock = resblock
+    self.resblock_kernel_sizes = resblock_kernel_sizes
+    self.resblock_dilation_sizes = resblock_dilation_sizes
+    self.upsample_rates = upsample_rates
+    self.upsample_initial_channel = upsample_initial_channel
+    self.upsample_kernel_sizes = upsample_kernel_sizes
+    self.segment_size = segment_size
+    self.gin_channels = gin_channels
+    self.ssl_dim = ssl_dim
+    self.emb_g = nn.Embedding(n_speakers, gin_channels)
+
+    self.enc_p_ = TextEncoder(ssl_dim, inter_channels, hidden_channels, 5, 1, 16,0, filter_channels, n_heads, p_dropout)
+    hps = {
+        "sampling_rate": 48000,
+        "inter_channels": 192,
+        "resblock": "1",
+        "resblock_kernel_sizes": [3, 7, 11],
+        "resblock_dilation_sizes": [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+        "upsample_rates": [10, 8, 2, 2],
+        "upsample_initial_channel": 512,
+        "upsample_kernel_sizes": [16, 16, 4, 4],
+        "gin_channels": 256,
+    }
+    self.dec = Generator(h=hps)
+    self.enc_q = Encoder(spec_channels, inter_channels, hidden_channels, 5, 1, 16, gin_channels=gin_channels)
+    self.flow = ResidualCouplingBlock(inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels)
+
+  def forward(self, c, f0, spec, g=None, mel=None, c_lengths=None, spec_lengths=None):
+    if c_lengths == None:
+      c_lengths = (torch.ones(c.size(0)) * c.size(-1)).to(c.device)
+    if spec_lengths == None:
+      spec_lengths = (torch.ones(spec.size(0)) * spec.size(-1)).to(spec.device)
+
+    g = self.emb_g(g).transpose(1,2)
+
+    z_ptemp, m_p, logs_p, _ = self.enc_p_(c, c_lengths, f0=f0_to_coarse(f0))
+    z, m_q, logs_q, spec_mask = self.enc_q(spec, spec_lengths, g=g) 
+
+    z_p = self.flow(z, spec_mask, g=g)
+    z_slice, pitch_slice, ids_slice = commons.rand_slice_segments_with_pitch(z, f0, spec_lengths, self.segment_size)
+
+    # o = self.dec(z_slice, g=g)
+    o = self.dec(z_slice, g=g, f0=pitch_slice)
+
+    return o, ids_slice, spec_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+  def infer(self, c, f0, g=None, mel=None, c_lengths=None):
+    if c_lengths == None:
+      c_lengths = (torch.ones(c.size(0)) * c.size(-1)).to(c.device)
+    g = self.emb_g(g).transpose(1,2)
+
+    z_p, m_p, logs_p, c_mask = self.enc_p_(c, c_lengths, f0=f0_to_coarse(f0))
+    z = self.flow(z_p, c_mask, g=g, reverse=True)
+
+    o = self.dec(z * c_mask, g=g, f0=f0)
+
+    return o

modules.py

@@ -0,0 +1,342 @@
+import copy
+import math
+import numpy as np
+import scipy
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+import commons
+from commons import init_weights, get_padding
+
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+  def __init__(self, channels, eps=1e-5):
+    super().__init__()
+    self.channels = channels
+    self.eps = eps
+
+    self.gamma = nn.Parameter(torch.ones(channels))
+    self.beta = nn.Parameter(torch.zeros(channels))
+
+  def forward(self, x):
+    x = x.transpose(1, -1)
+    x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+    return x.transpose(1, -1)
+
+ 
+class ConvReluNorm(nn.Module):
+  def __init__(self, in_channels, hidden_channels, out_channels, kernel_size, n_layers, p_dropout):
+    super().__init__()
+    self.in_channels = in_channels
+    self.hidden_channels = hidden_channels
+    self.out_channels = out_channels
+    self.kernel_size = kernel_size
+    self.n_layers = n_layers
+    self.p_dropout = p_dropout
+    assert n_layers > 1, "Number of layers should be larger than 0."
+
+    self.conv_layers = nn.ModuleList()
+    self.norm_layers = nn.ModuleList()
+    self.conv_layers.append(nn.Conv1d(in_channels, hidden_channels, kernel_size, padding=kernel_size//2))
+    self.norm_layers.append(LayerNorm(hidden_channels))
+    self.relu_drop = nn.Sequential(
+        nn.ReLU(),
+        nn.Dropout(p_dropout))
+    for _ in range(n_layers-1):
+      self.conv_layers.append(nn.Conv1d(hidden_channels, hidden_channels, kernel_size, padding=kernel_size//2))
+      self.norm_layers.append(LayerNorm(hidden_channels))
+    self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+    self.proj.weight.data.zero_()
+    self.proj.bias.data.zero_()
+
+  def forward(self, x, x_mask):
+    x_org = x
+    for i in range(self.n_layers):
+      x = self.conv_layers[i](x * x_mask)
+      x = self.norm_layers[i](x)
+      x = self.relu_drop(x)
+    x = x_org + self.proj(x)
+    return x * x_mask
+
+
+class DDSConv(nn.Module):
+  """
+  Dialted and Depth-Separable Convolution
+  """
+  def __init__(self, channels, kernel_size, n_layers, p_dropout=0.):
+    super().__init__()
+    self.channels = channels
+    self.kernel_size = kernel_size
+    self.n_layers = n_layers
+    self.p_dropout = p_dropout
+
+    self.drop = nn.Dropout(p_dropout)
+    self.convs_sep = nn.ModuleList()
+    self.convs_1x1 = nn.ModuleList()
+    self.norms_1 = nn.ModuleList()
+    self.norms_2 = nn.ModuleList()
+    for i in range(n_layers):
+      dilation = kernel_size ** i
+      padding = (kernel_size * dilation - dilation) // 2
+      self.convs_sep.append(nn.Conv1d(channels, channels, kernel_size, 
+          groups=channels, dilation=dilation, padding=padding
+      ))
+      self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+      self.norms_1.append(LayerNorm(channels))
+      self.norms_2.append(LayerNorm(channels))
+
+  def forward(self, x, x_mask, g=None):
+    if g is not None:
+      x = x + g
+    for i in range(self.n_layers):
+      y = self.convs_sep[i](x * x_mask)
+      y = self.norms_1[i](y)
+      y = F.gelu(y)
+      y = self.convs_1x1[i](y)
+      y = self.norms_2[i](y)
+      y = F.gelu(y)
+      y = self.drop(y)
+      x = x + y
+    return x * x_mask
+
+
+class WN(torch.nn.Module):
+  def __init__(self, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0, p_dropout=0):
+    super(WN, self).__init__()
+    assert(kernel_size % 2 == 1)
+    self.hidden_channels =hidden_channels
+    self.kernel_size = kernel_size,
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.gin_channels = gin_channels
+    self.p_dropout = p_dropout
+
+    self.in_layers = torch.nn.ModuleList()
+    self.res_skip_layers = torch.nn.ModuleList()
+    self.drop = nn.Dropout(p_dropout)
+
+    if gin_channels != 0:
+      cond_layer = torch.nn.Conv1d(gin_channels, 2*hidden_channels*n_layers, 1)
+      self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name='weight')
+
+    for i in range(n_layers):
+      dilation = dilation_rate ** i
+      padding = int((kernel_size * dilation - dilation) / 2)
+      in_layer = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, kernel_size,
+                                 dilation=dilation, padding=padding)
+      in_layer = torch.nn.utils.weight_norm(in_layer, name='weight')
+      self.in_layers.append(in_layer)
+
+      # last one is not necessary
+      if i < n_layers - 1:
+        res_skip_channels = 2 * hidden_channels
+      else:
+        res_skip_channels = hidden_channels
+
+      res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+      res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name='weight')
+      self.res_skip_layers.append(res_skip_layer)
+
+  def forward(self, x, x_mask, g=None, **kwargs):
+    output = torch.zeros_like(x)
+    n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+    if g is not None:
+      g = self.cond_layer(g)
+
+    for i in range(self.n_layers):
+      x_in = self.in_layers[i](x)
+      if g is not None:
+        cond_offset = i * 2 * self.hidden_channels
+        g_l = g[:,cond_offset:cond_offset+2*self.hidden_channels,:]
+      else:
+        g_l = torch.zeros_like(x_in)
+
+      acts = commons.fused_add_tanh_sigmoid_multiply(
+          x_in,
+          g_l,
+          n_channels_tensor)
+      acts = self.drop(acts)
+
+      res_skip_acts = self.res_skip_layers[i](acts)
+      if i < self.n_layers - 1:
+        res_acts = res_skip_acts[:,:self.hidden_channels,:]
+        x = (x + res_acts) * x_mask
+        output = output + res_skip_acts[:,self.hidden_channels:,:]
+      else:
+        output = output + res_skip_acts
+    return output * x_mask
+
+  def remove_weight_norm(self):
+    if self.gin_channels != 0:
+      torch.nn.utils.remove_weight_norm(self.cond_layer)
+    for l in self.in_layers:
+      torch.nn.utils.remove_weight_norm(l)
+    for l in self.res_skip_layers:
+     torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[2],
+                               padding=get_padding(kernel_size, dilation[2])))
+        ])
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1)))
+        ])
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1])))
+        ])
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+  def forward(self, x, x_mask, reverse=False, **kwargs):
+    if not reverse:
+      y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+      logdet = torch.sum(-y, [1, 2])
+      return y, logdet
+    else:
+      x = torch.exp(x) * x_mask
+      return x
+    
+
+class Flip(nn.Module):
+  def forward(self, x, *args, reverse=False, **kwargs):
+    x = torch.flip(x, [1])
+    if not reverse:
+      logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+      return x, logdet
+    else:
+      return x
+
+
+class ElementwiseAffine(nn.Module):
+  def __init__(self, channels):
+    super().__init__()
+    self.channels = channels
+    self.m = nn.Parameter(torch.zeros(channels,1))
+    self.logs = nn.Parameter(torch.zeros(channels,1))
+
+  def forward(self, x, x_mask, reverse=False, **kwargs):
+    if not reverse:
+      y = self.m + torch.exp(self.logs) * x
+      y = y * x_mask
+      logdet = torch.sum(self.logs * x_mask, [1,2])
+      return y, logdet
+    else:
+      x = (x - self.m) * torch.exp(-self.logs) * x_mask
+      return x
+
+
+class ResidualCouplingLayer(nn.Module):
+  def __init__(self,
+      channels,
+      hidden_channels,
+      kernel_size,
+      dilation_rate,
+      n_layers,
+      p_dropout=0,
+      gin_channels=0,
+      mean_only=False):
+    assert channels % 2 == 0, "channels should be divisible by 2"
+    super().__init__()
+    self.channels = channels
+    self.hidden_channels = hidden_channels
+    self.kernel_size = kernel_size
+    self.dilation_rate = dilation_rate
+    self.n_layers = n_layers
+    self.half_channels = channels // 2
+    self.mean_only = mean_only
+
+    self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+    self.enc = WN(hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=p_dropout, gin_channels=gin_channels)
+    self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+    self.post.weight.data.zero_()
+    self.post.bias.data.zero_()
+
+  def forward(self, x, x_mask, g=None, reverse=False):
+    x0, x1 = torch.split(x, [self.half_channels]*2, 1)
+    h = self.pre(x0) * x_mask
+    h = self.enc(h, x_mask, g=g)
+    stats = self.post(h) * x_mask
+    if not self.mean_only:
+      m, logs = torch.split(stats, [self.half_channels]*2, 1)
+    else:
+      m = stats
+      logs = torch.zeros_like(m)
+
+    if not reverse:
+      x1 = m + x1 * torch.exp(logs) * x_mask
+      x = torch.cat([x0, x1], 1)
+      logdet = torch.sum(logs, [1,2])
+      return x, logdet
+    else:
+      x1 = (x1 - m) * torch.exp(-logs) * x_mask
+      x = torch.cat([x0, x1], 1)
+      return x

raw/大手拉小手.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b89788bc5dfeb7a9662665a8c07f4a7f49c74e245282b7a1b2f21efa07ad121c
+size 2808078

requirements.txt

@@ -0,0 +1,16 @@
+numpy
+playsound
+pydub
+pyworld
+requests
+scipy
+sounddevice
+SoundFile
+starlette
+torch
+torchaudio
+tqdm
+scikit-maad
+praat-parselmouth
+librosa
+torchvision

spec_gen.py

@@ -0,0 +1,22 @@
+from data_utils import TextAudioSpeakerLoader, EvalDataLoader
+import json
+from tqdm import tqdm
+
+from utils import HParams
+
+config_path = 'configs/config.json'
+with open(config_path, "r") as f:
+    data = f.read()
+config = json.loads(data)
+hps = HParams(**config)
+
+train_dataset = TextAudioSpeakerLoader("filelists/train.txt", hps)
+test_dataset = TextAudioSpeakerLoader("filelists/test.txt", hps)
+eval_dataset = TextAudioSpeakerLoader("filelists/val.txt", hps)
+
+for _ in tqdm(train_dataset):
+    pass
+for _ in tqdm(eval_dataset):
+    pass
+for _ in tqdm(test_dataset):
+    pass

terms.md

@@ -0,0 +1,65 @@
+# 在使用此模型前请阅读以下协议
+
+## AI阿夸模型使用协议
+
+**【前言】**AI阿夸模型所有者及训练者@MasterSatori（以下也称“我”）希望通过《AI阿夸模型使用协议》（以下简称“本协议”）向您说明您在使用AI阿夸模型时应当履行的责任及使用范围。
+
+**【特别提示】**在使用AI阿夸模型前，请您务必仔细阅读并透彻理解本协议，特别是以粗体标识的条款，您应重点阅读，在确认充分理解并同意后再开始使用。
+
+​	**本协议将帮助您了解以下内容：**
+
+​	**一、免责声明**
+
+​	**二、您在非个人使用场合时使用AI阿夸模型应当做的事**
+
+​	**三、AI阿夸模型的使用范围**
+
+​	**四、如何联系我**
+
+​	**(一) 免责声明：**
+
+​	**您因使用AI阿夸模型对其它任何实体（个人/企业）所造成的任何损失由您自身承担，您因使用AI阿夸模型所产生的一切法律风险及法律纠纷由您自身承担。**
+
+​	**(二) 您在非个人使用场合时使用AI阿夸模型应当做的事：**
+
+​	1、注明soVITS项目作者：Rcell
+
+​	2、注明我：MasterSatori
+
+​	**(三) AI阿夸模型的使用范围：**
+
+​	1、您可以使用的范围：
+
+​	(1) 个人使用√
+
+​	(2) 将产生的音频用于投稿（投稿内容不得包含“您不可使用的范围”中的内容）√
+
+​	(3) 二创内容√
+
+​	(4) 非付费粉丝向同人游戏√
+
+​	2、您不可使用的范围：
+
+​	(1) 商业使用×
+
+​	(2) 假冒本人×
+
+​	(3) 当作变声器等使用（个人使用除外）×
+
+​	(4) 将AI阿夸模型再次上传×
+
+​	(5) 将AI阿夸模型作为底模进行训练×
+
+​	(6) 低创内容（音频中有过多的爆音或电音属于“低创内容”）×
+
+​	(7) 敏感内容×
+
+​	3、补充内容：
+
+​	在其他未被提及的场合使用AI阿夸模型及其所产生的数据您应当联系我给出答复。
+
+​	**（四）如何联系我：**
+
+​	1、可在b站私信我（UID:8407182）
+
+​	2、可发送邮件至：[email protected]

utils.py

@@ -0,0 +1,338 @@
+import os
+import glob
+import sys
+import argparse
+import logging
+import json
+import subprocess
+
+import librosa
+import numpy as np
+import torchaudio
+from scipy.io.wavfile import read
+import torch
+import torchvision
+from torch.nn import functional as F
+from commons import sequence_mask
+from hubert import hubert_model
+MATPLOTLIB_FLAG = False
+
+logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
+logger = logging
+
+f0_bin = 256
+f0_max = 1100.0
+f0_min = 50.0
+f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+
+def f0_to_coarse(f0):
+  is_torch = isinstance(f0, torch.Tensor)
+  f0_mel = 1127 * (1 + f0 / 700).log() if is_torch else 1127 * np.log(1 + f0 / 700)
+  f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * (f0_bin - 2) / (f0_mel_max - f0_mel_min) + 1
+
+  f0_mel[f0_mel <= 1] = 1
+  f0_mel[f0_mel > f0_bin - 1] = f0_bin - 1
+  f0_coarse = (f0_mel + 0.5).long() if is_torch else np.rint(f0_mel).astype(np.int)
+  assert f0_coarse.max() <= 255 and f0_coarse.min() >= 1, (f0_coarse.max(), f0_coarse.min())
+  return f0_coarse
+
+
+def get_hubert_model(rank=None):
+
+  hubert_soft = hubert_model.hubert_soft("hubert/hubert-soft-0d54a1f4.pt")
+  if rank is not None:
+    hubert_soft = hubert_soft.cuda(rank)
+  return hubert_soft
+
+def get_hubert_content(hmodel, y=None, path=None):
+  if path is not None:
+    source, sr = torchaudio.load(path)
+    source = torchaudio.functional.resample(source, sr, 16000)
+    if len(source.shape) == 2 and source.shape[1] >= 2:
+      source = torch.mean(source, dim=0).unsqueeze(0)
+  else:
+    source = y
+  source = source.unsqueeze(0)
+  with torch.inference_mode():
+    units = hmodel.units(source)
+    return units.transpose(1,2)
+
+
+def get_content(cmodel, y):
+    with torch.no_grad():
+        c = cmodel.extract_features(y.squeeze(1))[0]
+    c = c.transpose(1, 2)
+    return c
+
+
+
+def transform(mel, height): # 68-92
+    #r = np.random.random()
+    #rate = r * 0.3 + 0.85 # 0.85-1.15
+    #height = int(mel.size(-2) * rate)
+    tgt = torchvision.transforms.functional.resize(mel, (height, mel.size(-1)))
+    if height >= mel.size(-2):
+        return tgt[:, :mel.size(-2), :]
+    else:
+        silence = tgt[:,-1:,:].repeat(1,mel.size(-2)-height,1)
+        silence += torch.randn_like(silence) / 10
+        return torch.cat((tgt, silence), 1)
+
+
+def stretch(mel, width): # 0.5-2
+    return torchvision.transforms.functional.resize(mel, (mel.size(-2), width))
+
+
+def load_checkpoint(checkpoint_path, model, optimizer=None):
+  assert os.path.isfile(checkpoint_path)
+  checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
+  iteration = checkpoint_dict['iteration']
+  learning_rate = checkpoint_dict['learning_rate']
+  if iteration is None:
+    iteration = 1
+  if learning_rate is None:
+    learning_rate = 0.0002
+  if optimizer is not None and checkpoint_dict['optimizer'] is not None:
+    optimizer.load_state_dict(checkpoint_dict['optimizer'])
+  saved_state_dict = checkpoint_dict['model']
+  if hasattr(model, 'module'):
+    state_dict = model.module.state_dict()
+  else:
+    state_dict = model.state_dict()
+  new_state_dict= {}
+  for k, v in state_dict.items():
+    try:
+      new_state_dict[k] = saved_state_dict[k]
+    except:
+      logger.info("%s is not in the checkpoint" % k)
+      new_state_dict[k] = v
+  if hasattr(model, 'module'):
+    model.module.load_state_dict(new_state_dict)
+  else:
+    model.load_state_dict(new_state_dict)
+  logger.info("Loaded checkpoint '{}' (iteration {})" .format(
+    checkpoint_path, iteration))
+  return model, optimizer, learning_rate, iteration
+
+
+def save_checkpoint(model, optimizer, learning_rate, iteration, checkpoint_path):
+  # ckptname = checkpoint_path.split(os.sep)[-1]
+  # newest_step = int(ckptname.split(".")[0].split("_")[1])
+  # val_steps = 2000
+  # last_ckptname = checkpoint_path.replace(str(newest_step), str(newest_step - val_steps*3))
+  # if newest_step >= val_steps*3:
+  #   os.system(f"rm {last_ckptname}")
+  logger.info("Saving model and optimizer state at iteration {} to {}".format(
+    iteration, checkpoint_path))
+  if hasattr(model, 'module'):
+    state_dict = model.module.state_dict()
+  else:
+    state_dict = model.state_dict()
+  torch.save({'model': state_dict,
+              'iteration': iteration,
+              'optimizer': optimizer.state_dict(),
+              'learning_rate': learning_rate}, checkpoint_path)
+
+
+def summarize(writer, global_step, scalars={}, histograms={}, images={}, audios={}, audio_sampling_rate=22050):
+  for k, v in scalars.items():
+    writer.add_scalar(k, v, global_step)
+  for k, v in histograms.items():
+    writer.add_histogram(k, v, global_step)
+  for k, v in images.items():
+    writer.add_image(k, v, global_step, dataformats='HWC')
+  for k, v in audios.items():
+    writer.add_audio(k, v, global_step, audio_sampling_rate)
+
+
+def latest_checkpoint_path(dir_path, regex="G_*.pth"):
+  f_list = glob.glob(os.path.join(dir_path, regex))
+  f_list.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
+  x = f_list[-1]
+  print(x)
+  return x
+
+
+def plot_spectrogram_to_numpy(spectrogram):
+  global MATPLOTLIB_FLAG
+  if not MATPLOTLIB_FLAG:
+    import matplotlib
+    matplotlib.use("Agg")
+    MATPLOTLIB_FLAG = True
+    mpl_logger = logging.getLogger('matplotlib')
+    mpl_logger.setLevel(logging.WARNING)
+  import matplotlib.pylab as plt
+  import numpy as np
+
+  fig, ax = plt.subplots(figsize=(10,2))
+  im = ax.imshow(spectrogram, aspect="auto", origin="lower",
+                  interpolation='none')
+  plt.colorbar(im, ax=ax)
+  plt.xlabel("Frames")
+  plt.ylabel("Channels")
+  plt.tight_layout()
+
+  fig.canvas.draw()
+  data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+  data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+  plt.close()
+  return data
+
+
+def plot_alignment_to_numpy(alignment, info=None):
+  global MATPLOTLIB_FLAG
+  if not MATPLOTLIB_FLAG:
+    import matplotlib
+    matplotlib.use("Agg")
+    MATPLOTLIB_FLAG = True
+    mpl_logger = logging.getLogger('matplotlib')
+    mpl_logger.setLevel(logging.WARNING)
+  import matplotlib.pylab as plt
+  import numpy as np
+
+  fig, ax = plt.subplots(figsize=(6, 4))
+  im = ax.imshow(alignment.transpose(), aspect='auto', origin='lower',
+                  interpolation='none')
+  fig.colorbar(im, ax=ax)
+  xlabel = 'Decoder timestep'
+  if info is not None:
+      xlabel += '\n\n' + info
+  plt.xlabel(xlabel)
+  plt.ylabel('Encoder timestep')
+  plt.tight_layout()
+
+  fig.canvas.draw()
+  data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')
+  data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+  plt.close()
+  return data
+
+
+def load_wav_to_torch(full_path):
+  sampling_rate, data = read(full_path)
+  return torch.FloatTensor(data.astype(np.float32)), sampling_rate
+
+
+def load_filepaths_and_text(filename, split="|"):
+  with open(filename, encoding='utf-8') as f:
+    filepaths_and_text = [line.strip().split(split) for line in f]
+  return filepaths_and_text
+
+
+def get_hparams(init=True):
+  parser = argparse.ArgumentParser()
+  parser.add_argument('-c', '--config', type=str, default="./configs/base.json",
+                      help='JSON file for configuration')
+  parser.add_argument('-m', '--model', type=str, required=True,
+                      help='Model name')
+
+  args = parser.parse_args()
+  model_dir = os.path.join("./logs", args.model)
+
+  if not os.path.exists(model_dir):
+    os.makedirs(model_dir)
+
+  config_path = args.config
+  config_save_path = os.path.join(model_dir, "config.json")
+  if init:
+    with open(config_path, "r") as f:
+      data = f.read()
+    with open(config_save_path, "w") as f:
+      f.write(data)
+  else:
+    with open(config_save_path, "r") as f:
+      data = f.read()
+  config = json.loads(data)
+
+  hparams = HParams(**config)
+  hparams.model_dir = model_dir
+  return hparams
+
+
+def get_hparams_from_dir(model_dir):
+  config_save_path = os.path.join(model_dir, "config.json")
+  with open(config_save_path, "r") as f:
+    data = f.read()
+  config = json.loads(data)
+
+  hparams =HParams(**config)
+  hparams.model_dir = model_dir
+  return hparams
+
+
+def get_hparams_from_file(config_path):
+  with open(config_path, "r") as f:
+    data = f.read()
+  config = json.loads(data)
+
+  hparams =HParams(**config)
+  return hparams
+
+
+def check_git_hash(model_dir):
+  source_dir = os.path.dirname(os.path.realpath(__file__))
+  if not os.path.exists(os.path.join(source_dir, ".git")):
+    logger.warn("{} is not a git repository, therefore hash value comparison will be ignored.".format(
+      source_dir
+    ))
+    return
+
+  cur_hash = subprocess.getoutput("git rev-parse HEAD")
+
+  path = os.path.join(model_dir, "githash")
+  if os.path.exists(path):
+    saved_hash = open(path).read()
+    if saved_hash != cur_hash:
+      logger.warn("git hash values are different. {}(saved) != {}(current)".format(
+        saved_hash[:8], cur_hash[:8]))
+  else:
+    open(path, "w").write(cur_hash)
+
+
+def get_logger(model_dir, filename="train.log"):
+  global logger
+  logger = logging.getLogger(os.path.basename(model_dir))
+  logger.setLevel(logging.DEBUG)
+
+  formatter = logging.Formatter("%(asctime)s\t%(name)s\t%(levelname)s\t%(message)s")
+  if not os.path.exists(model_dir):
+    os.makedirs(model_dir)
+  h = logging.FileHandler(os.path.join(model_dir, filename))
+  h.setLevel(logging.DEBUG)
+  h.setFormatter(formatter)
+  logger.addHandler(h)
+  return logger
+
+
+class HParams():
+  def __init__(self, **kwargs):
+    for k, v in kwargs.items():
+      if type(v) == dict:
+        v = HParams(**v)
+      self[k] = v
+
+  def keys(self):
+    return self.__dict__.keys()
+
+  def items(self):
+    return self.__dict__.items()
+
+  def values(self):
+    return self.__dict__.values()
+
+  def __len__(self):
+    return len(self.__dict__)
+
+  def __getitem__(self, key):
+    return getattr(self, key)
+
+  def __setitem__(self, key, value):
+    return setattr(self, key, value)
+
+  def __contains__(self, key):
+    return key in self.__dict__
+
+  def __repr__(self):
+    return self.__dict__.__repr__()
+

vdecoder/hifigan/env.py

@@ -0,0 +1,15 @@
+import os
+import shutil
+
+
+class AttrDict(dict):
+    def __init__(self, *args, **kwargs):
+        super(AttrDict, self).__init__(*args, **kwargs)
+        self.__dict__ = self
+
+
+def build_env(config, config_name, path):
+    t_path = os.path.join(path, config_name)
+    if config != t_path:
+        os.makedirs(path, exist_ok=True)
+        shutil.copyfile(config, os.path.join(path, config_name))

vdecoder/hifigan/models.py

@@ -0,0 +1,500 @@
+import os
+import json
+from .env import AttrDict
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.nn as nn
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from .utils import init_weights, get_padding
+
+LRELU_SLOPE = 0.1
+
+
+def load_model(model_path, device='cuda'):
+    config_file = os.path.join(os.path.split(model_path)[0], 'config.json')
+    with open(config_file) as f:
+        data = f.read()
+
+    global h
+    json_config = json.loads(data)
+    h = AttrDict(json_config)
+
+    generator = Generator(h).to(device)
+
+    cp_dict = torch.load(model_path)
+    generator.load_state_dict(cp_dict['generator'])
+    generator.eval()
+    generator.remove_weight_norm()
+    del cp_dict
+    return generator, h
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.h = h
+        self.convs1 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[2],
+                               padding=get_padding(kernel_size, dilation[2])))
+        ])
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
+                               padding=get_padding(kernel_size, 1)))
+        ])
+        self.convs2.apply(init_weights)
+
+    def forward(self, x):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            xt = c2(xt)
+            x = xt + x
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.h = h
+        self.convs = nn.ModuleList([
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
+                               padding=get_padding(kernel_size, dilation[0]))),
+            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
+                               padding=get_padding(kernel_size, dilation[1])))
+        ])
+        self.convs.apply(init_weights)
+
+    def forward(self, x):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            xt = c(xt)
+            x = xt + x
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class SineGen(torch.nn.Module):
+    """ Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(self, samp_rate, harmonic_num=0,
+                 sine_amp=0.1, noise_std=0.003,
+                 voiced_threshold=0,
+                 flag_for_pulse=False):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+        self.flag_for_pulse = flag_for_pulse
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = (f0 > self.voiced_threshold).type(torch.float32)
+        return uv
+
+    def _f02sine(self, f0_values):
+        """ f0_values: (batchsize, length, dim)
+            where dim indicates fundamental tone and overtones
+        """
+        # convert to F0 in rad. The interger part n can be ignored
+        # because 2 * np.pi * n doesn't affect phase
+        rad_values = (f0_values / self.sampling_rate) % 1
+
+        # initial phase noise (no noise for fundamental component)
+        rand_ini = torch.rand(f0_values.shape[0], f0_values.shape[2], \
+                              device=f0_values.device)
+        rand_ini[:, 0] = 0
+        rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+
+        # instantanouse phase sine[t] = sin(2*pi \sum_i=1 ^{t} rad)
+        if not self.flag_for_pulse:
+            # for normal case
+
+            # To prevent torch.cumsum numerical overflow,
+            # it is necessary to add -1 whenever \sum_k=1^n rad_value_k > 1.
+            # Buffer tmp_over_one_idx indicates the time step to add -1.
+            # This will not change F0 of sine because (x-1) * 2*pi = x * 2*pi
+            tmp_over_one = torch.cumsum(rad_values, 1) % 1
+            tmp_over_one_idx = (torch.diff(tmp_over_one, dim=1)) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+
+            sines = torch.sin(torch.cumsum(rad_values + cumsum_shift, dim=1)
+                              * 2 * np.pi)
+        else:
+            # If necessary, make sure that the first time step of every
+            # voiced segments is sin(pi) or cos(0)
+            # This is used for pulse-train generation
+
+            # identify the last time step in unvoiced segments
+            uv = self._f02uv(f0_values)
+            uv_1 = torch.roll(uv, shifts=-1, dims=1)
+            uv_1[:, -1, :] = 1
+            u_loc = (uv < 1) * (uv_1 > 0)
+
+            # get the instantanouse phase
+            tmp_cumsum = torch.cumsum(rad_values, dim=1)
+            # different batch needs to be processed differently
+            for idx in range(f0_values.shape[0]):
+                temp_sum = tmp_cumsum[idx, u_loc[idx, :, 0], :]
+                temp_sum[1:, :] = temp_sum[1:, :] - temp_sum[0:-1, :]
+                # stores the accumulation of i.phase within
+                # each voiced segments
+                tmp_cumsum[idx, :, :] = 0
+                tmp_cumsum[idx, u_loc[idx, :, 0], :] = temp_sum
+
+            # rad_values - tmp_cumsum: remove the accumulation of i.phase
+            # within the previous voiced segment.
+            i_phase = torch.cumsum(rad_values - tmp_cumsum, dim=1)
+
+            # get the sines
+            sines = torch.cos(i_phase * 2 * np.pi)
+        return sines
+
+    def forward(self, f0):
+        """ sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim,
+                                 device=f0.device)
+            # fundamental component
+            fn = torch.multiply(f0, torch.FloatTensor([[range(1, self.harmonic_num + 2)]]).to(f0.device))
+
+            # generate sine waveforms
+            sine_waves = self._f02sine(fn) * self.sine_amp
+
+            # generate uv signal
+            # uv = torch.ones(f0.shape)
+            # uv = uv * (f0 > self.voiced_threshold)
+            uv = self._f02uv(f0)
+
+            # noise: for unvoiced should be similar to sine_amp
+            #        std = self.sine_amp/3 -> max value ~ self.sine_amp
+            # .       for voiced regions is self.noise_std
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+
+            # first: set the unvoiced part to 0 by uv
+            # then: additive noise
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """ SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(self, sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(sampling_rate, harmonic_num,
+                                 sine_amp, add_noise_std, voiced_threshod)
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x):
+        """
+        Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+        F0_sampled (batchsize, length, 1)
+        Sine_source (batchsize, length, 1)
+        noise_source (batchsize, length 1)
+        """
+        # source for harmonic branch
+        sine_wavs, uv, _ = self.l_sin_gen(x)
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+
+        # source for noise branch, in the same shape as uv
+        noise = torch.randn_like(uv) * self.sine_amp / 3
+        return sine_merge, noise, uv
+
+
+class Generator(torch.nn.Module):
+    def __init__(self, h):
+        super(Generator, self).__init__()
+        self.h = h
+
+        self.num_kernels = len(h["resblock_kernel_sizes"])
+        self.num_upsamples = len(h["upsample_rates"])
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(h["upsample_rates"]))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=h["sampling_rate"],
+            harmonic_num=8)
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = weight_norm(Conv1d(h["inter_channels"], h["upsample_initial_channel"], 7, 1, padding=3))
+        resblock = ResBlock1 if h["resblock"] == '1' else ResBlock2
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(h["upsample_rates"], h["upsample_kernel_sizes"])):
+            c_cur = h["upsample_initial_channel"] // (2 ** (i + 1))
+            self.ups.append(weight_norm(
+                ConvTranspose1d(h["upsample_initial_channel"] // (2 ** i), h["upsample_initial_channel"] // (2 ** (i + 1)),
+                                k, u, padding=(k - u) // 2)))
+            if i + 1 < len(h["upsample_rates"]):  #
+                stride_f0 = np.prod(h["upsample_rates"][i + 1:])
+                self.noise_convs.append(Conv1d(
+                    1, c_cur, kernel_size=stride_f0 * 2, stride=stride_f0, padding=stride_f0 // 2))
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = h["upsample_initial_channel"] // (2 ** (i + 1))
+            for j, (k, d) in enumerate(zip(h["resblock_kernel_sizes"], h["resblock_dilation_sizes"])):
+                self.resblocks.append(resblock(h, ch, k, d))
+
+        self.conv_post = weight_norm(Conv1d(ch, 1, 7, 1, padding=3))
+        self.ups.apply(init_weights)
+        self.conv_post.apply(init_weights)
+        self.cond = nn.Conv1d(h['gin_channels'], h['upsample_initial_channel'], 1)
+
+    def forward(self, x, f0, g=None):
+        # print(1,x.shape,f0.shape,f0[:, None].shape)
+        f0 = self.f0_upsamp(f0[:, None]).transpose(1, 2)  # bs,n,t
+        # print(2,f0.shape)
+        har_source, noi_source, uv = self.m_source(f0)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        x = x + self.cond(g)
+        # print(124,x.shape,har_source.shape)
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, LRELU_SLOPE)
+            # print(3,x.shape)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            # print(4,x_source.shape,har_source.shape,x.shape)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        print('Removing weight norm...')
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+        remove_weight_norm(self.conv_pre)
+        remove_weight_norm(self.conv_post)
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
+            norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
+            norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
+            norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
+            norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(2, 0))),
+        ])
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, periods=None):
+        super(MultiPeriodDiscriminator, self).__init__()
+        self.periods = periods if periods is not None else [2, 3, 5, 7, 11]
+        self.discriminators = nn.ModuleList()
+        for period in self.periods:
+            self.discriminators.append(DiscriminatorP(period))
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            fmap_rs.append(fmap_r)
+            y_d_gs.append(y_d_g)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList([
+            norm_f(Conv1d(1, 128, 15, 1, padding=7)),
+            norm_f(Conv1d(128, 128, 41, 2, groups=4, padding=20)),
+            norm_f(Conv1d(128, 256, 41, 2, groups=16, padding=20)),
+            norm_f(Conv1d(256, 512, 41, 4, groups=16, padding=20)),
+            norm_f(Conv1d(512, 1024, 41, 4, groups=16, padding=20)),
+            norm_f(Conv1d(1024, 1024, 41, 1, groups=16, padding=20)),
+            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+        ])
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiScaleDiscriminator(torch.nn.Module):
+    def __init__(self):
+        super(MultiScaleDiscriminator, self).__init__()
+        self.discriminators = nn.ModuleList([
+            DiscriminatorS(use_spectral_norm=True),
+            DiscriminatorS(),
+            DiscriminatorS(),
+        ])
+        self.meanpools = nn.ModuleList([
+            AvgPool1d(4, 2, padding=2),
+            AvgPool1d(4, 2, padding=2)
+        ])
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            if i != 0:
+                y = self.meanpools[i - 1](y)
+                y_hat = self.meanpools[i - 1](y_hat)
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            fmap_rs.append(fmap_r)
+            y_d_gs.append(y_d_g)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+def feature_loss(fmap_r, fmap_g):
+    loss = 0
+    for dr, dg in zip(fmap_r, fmap_g):
+        for rl, gl in zip(dr, dg):
+            loss += torch.mean(torch.abs(rl - gl))
+
+    return loss * 2
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+    loss = 0
+    r_losses = []
+    g_losses = []
+    for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+        r_loss = torch.mean((1 - dr) ** 2)
+        g_loss = torch.mean(dg ** 2)
+        loss += (r_loss + g_loss)
+        r_losses.append(r_loss.item())
+        g_losses.append(g_loss.item())
+
+    return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+    loss = 0
+    gen_losses = []
+    for dg in disc_outputs:
+        l = torch.mean((1 - dg) ** 2)
+        gen_losses.append(l)
+        loss += l
+
+    return loss, gen_losses

vdecoder/hifigan/nvSTFT.py

@@ -0,0 +1,111 @@
+import math
+import os
+os.environ["LRU_CACHE_CAPACITY"] = "3"
+import random
+import torch
+import torch.utils.data
+import numpy as np
+import librosa
+from librosa.util import normalize
+from librosa.filters import mel as librosa_mel_fn
+from scipy.io.wavfile import read
+import soundfile as sf
+
+def load_wav_to_torch(full_path, target_sr=None, return_empty_on_exception=False):
+    sampling_rate = None
+    try:
+        data, sampling_rate = sf.read(full_path, always_2d=True)# than soundfile.
+    except Exception as ex:
+        print(f"'{full_path}' failed to load.\nException:")
+        print(ex)
+        if return_empty_on_exception:
+            return [], sampling_rate or target_sr or 48000
+        else:
+            raise Exception(ex)
+    
+    if len(data.shape) > 1:
+        data = data[:, 0]
+        assert len(data) > 2# check duration of audio file is > 2 samples (because otherwise the slice operation was on the wrong dimension)
+    
+    if np.issubdtype(data.dtype, np.integer): # if audio data is type int
+        max_mag = -np.iinfo(data.dtype).min # maximum magnitude = min possible value of intXX
+    else: # if audio data is type fp32
+        max_mag = max(np.amax(data), -np.amin(data))
+        max_mag = (2**31)+1 if max_mag > (2**15) else ((2**15)+1 if max_mag > 1.01 else 1.0) # data should be either 16-bit INT, 32-bit INT or [-1 to 1] float32
+    
+    data = torch.FloatTensor(data.astype(np.float32))/max_mag
+    
+    if (torch.isinf(data) | torch.isnan(data)).any() and return_empty_on_exception:# resample will crash with inf/NaN inputs. return_empty_on_exception will return empty arr instead of except
+        return [], sampling_rate or target_sr or 48000
+    if target_sr is not None and sampling_rate != target_sr:
+        data = torch.from_numpy(librosa.core.resample(data.numpy(), orig_sr=sampling_rate, target_sr=target_sr))
+        sampling_rate = target_sr
+    
+    return data, sampling_rate
+
+def dynamic_range_compression(x, C=1, clip_val=1e-5):
+    return np.log(np.clip(x, a_min=clip_val, a_max=None) * C)
+
+def dynamic_range_decompression(x, C=1):
+    return np.exp(x) / C
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+def dynamic_range_decompression_torch(x, C=1):
+    return torch.exp(x) / C
+
+class STFT():
+    def __init__(self, sr=22050, n_mels=80, n_fft=1024, win_size=1024, hop_length=256, fmin=20, fmax=11025, clip_val=1e-5):
+        self.target_sr = sr
+        
+        self.n_mels     = n_mels
+        self.n_fft      = n_fft
+        self.win_size   = win_size
+        self.hop_length = hop_length
+        self.fmin     = fmin
+        self.fmax     = fmax
+        self.clip_val = clip_val
+        self.mel_basis = {}
+        self.hann_window = {}
+    
+    def get_mel(self, y, center=False):
+        sampling_rate = self.target_sr
+        n_mels     = self.n_mels
+        n_fft      = self.n_fft
+        win_size   = self.win_size
+        hop_length = self.hop_length
+        fmin       = self.fmin
+        fmax       = self.fmax
+        clip_val   = self.clip_val
+        
+        if torch.min(y) < -1.:
+            print('min value is ', torch.min(y))
+        if torch.max(y) > 1.:
+            print('max value is ', torch.max(y))
+        
+        if fmax not in self.mel_basis:
+            mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=n_mels, fmin=fmin, fmax=fmax)
+            self.mel_basis[str(fmax)+'_'+str(y.device)] = torch.from_numpy(mel).float().to(y.device)
+            self.hann_window[str(y.device)] = torch.hann_window(self.win_size).to(y.device)
+        
+        y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_length)/2), int((n_fft-hop_length)/2)), mode='reflect')
+        y = y.squeeze(1)
+        
+        spec = torch.stft(y, n_fft, hop_length=hop_length, win_length=win_size, window=self.hann_window[str(y.device)],
+                          center=center, pad_mode='reflect', normalized=False, onesided=True)
+        # print(111,spec)
+        spec = torch.sqrt(spec.pow(2).sum(-1)+(1e-9))
+        # print(222,spec)
+        spec = torch.matmul(self.mel_basis[str(fmax)+'_'+str(y.device)], spec)
+        # print(333,spec)
+        spec = dynamic_range_compression_torch(spec, clip_val=clip_val)
+        # print(444,spec)
+        return spec
+    
+    def __call__(self, audiopath):
+        audio, sr = load_wav_to_torch(audiopath, target_sr=self.target_sr)
+        spect = self.get_mel(audio.unsqueeze(0)).squeeze(0)
+        return spect
+
+stft = STFT()

vdecoder/hifigan/utils.py

@@ -0,0 +1,68 @@
+import glob
+import os
+import matplotlib
+import torch
+from torch.nn.utils import weight_norm
+matplotlib.use("Agg")
+import matplotlib.pylab as plt
+
+
+def plot_spectrogram(spectrogram):
+    fig, ax = plt.subplots(figsize=(10, 2))
+    im = ax.imshow(spectrogram, aspect="auto", origin="lower",
+                   interpolation='none')
+    plt.colorbar(im, ax=ax)
+
+    fig.canvas.draw()
+    plt.close()
+
+    return fig
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def apply_weight_norm(m):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        weight_norm(m)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size*dilation - dilation)/2)
+
+
+def load_checkpoint(filepath, device):
+    assert os.path.isfile(filepath)
+    print("Loading '{}'".format(filepath))
+    checkpoint_dict = torch.load(filepath, map_location=device)
+    print("Complete.")
+    return checkpoint_dict
+
+
+def save_checkpoint(filepath, obj):
+    print("Saving checkpoint to {}".format(filepath))
+    torch.save(obj, filepath)
+    print("Complete.")
+
+
+def del_old_checkpoints(cp_dir, prefix, n_models=2):
+    pattern = os.path.join(cp_dir, prefix + '????????')
+    cp_list = glob.glob(pattern) # get checkpoint paths
+    cp_list = sorted(cp_list)# sort by iter
+    if len(cp_list) > n_models: # if more than n_models models are found
+        for cp in cp_list[:-n_models]:# delete the oldest models other than lastest n_models
+            open(cp, 'w').close()# empty file contents
+            os.unlink(cp)# delete file (move to trash when using Colab)
+
+
+def scan_checkpoint(cp_dir, prefix):
+    pattern = os.path.join(cp_dir, prefix + '????????')
+    cp_list = glob.glob(pattern)
+    if len(cp_list) == 0:
+        return None
+    return sorted(cp_list)[-1]
+