Spaces:
Running
on
Zero
Running
on
Zero
File size: 3,500 Bytes
8e8cd3e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 |
from models.mossformer2_se.mossformer2 import MossFormer_MaskNet
import torch.nn as nn
class MossFormer2_SE_48K(nn.Module):
"""
The MossFormer2_SE_48K model for speech enhancement.
This class encapsulates the functionality of the MossFormer MaskNet
within a higher-level model. It processes input audio data to produce
enhanced outputs and corresponding masks.
Arguments
---------
args : Namespace
Configuration arguments that may include hyperparameters
and model settings (not utilized in this implementation but
can be extended for flexibility).
Example
---------
>>> model = MossFormer2_SE_48K(args).model
>>> x = torch.randn(10, 180, 2000) # Example input
>>> outputs, mask = model(x) # Forward pass
>>> outputs.shape, mask.shape # Check output shapes
"""
def __init__(self, args):
super(MossFormer2_SE_48K, self).__init__()
# Initialize the TestNet model, which contains the MossFormer MaskNet
self.model = TestNet() # Instance of TestNet
def forward(self, x):
"""
Forward pass through the model.
Arguments
---------
x : torch.Tensor
Input tensor of dimension [B, N, S], where B is the batch size,
N is the number of channels (180 in this case), and S is the
sequence length (e.g., time frames).
Returns
-------
outputs : torch.Tensor
Enhanced audio output tensor from the model.
mask : torch.Tensor
Mask tensor predicted by the model for speech separation.
"""
outputs, mask = self.model(x) # Get outputs and mask from TestNet
return outputs, mask # Return the outputs and mask
class TestNet(nn.Module):
"""
The TestNet class for testing the MossFormer MaskNet implementation.
This class builds a model that integrates the MossFormer_MaskNet
for processing input audio and generating masks for source separation.
Arguments
---------
n_layers : int
The number of layers in the model. It determines the depth
of the model architecture, we leave this para unused at this moment.
"""
def __init__(self, n_layers=18):
super(TestNet, self).__init__()
self.n_layers = n_layers # Set the number of layers
# Initialize the MossFormer MaskNet with specified input and output channels
self.mossformer = MossFormer_MaskNet(in_channels=180, out_channels=512, out_channels_final=961)
def forward(self, input):
"""
Forward pass through the TestNet model.
Arguments
---------
input : torch.Tensor
Input tensor of dimension [B, N, S], where B is the batch size,
N is the number of input channels (180), and S is the sequence length.
Returns
-------
out_list : list
List containing the mask tensor predicted by the MossFormer_MaskNet.
"""
out_list = [] # Initialize output list to store outputs
# Transpose input to match expected shape for MaskNet
x = input.transpose(1, 2) # Change shape from [B, N, S] to [B, S, N]
# Get the mask from the MossFormer MaskNet
mask = self.mossformer(x) # Forward pass through the MossFormer_MaskNet
out_list.append(mask) # Append the mask to the output list
return out_list # Return the list containing the mask
|