ClearVoice / config /inference /MossFormerGAN_SE_16K.yaml
alibabasglab's picture
Upload 161 files
8e8cd3e verified
raw
history blame
828 Bytes
#!/bin/bash
mode: 'inference'
use_cuda: 1 # 1 for True, 0 for False
num_gpu: 1
sampling_rate: 16000
network: "MossFormerGAN_SE_16K" ##network type
checkpoint_dir: "checkpoints/MossFormerGAN_SE_16K"
#input_path: "data/cv_webrtc_test_set_20200521_16k.scp" # an input dir or input scp file
#input_path: "/home/shengkui.zhao/DingTalk_NS/data/webrtc_test_set_20200521_16k/noisy"
input_path: "/mnt/nas_sg/mit_sg/shengkui.zhao/ComplexNN/audio/youtube_testset_16k/noisy_long/noisy"
output_dir: "outputs/MossFormerGAN_SE_16K" ## output dir to store processed audio
# decode parameters
one_time_decode_length: 10 #maximum segment length for one-pass decoding (seconds), longer audio will use segmented decoding
decode_window: 10 #one-pass decoding length
# FFT parameters
win_type: 'hamming'
win_len: 400
win_inc: 100
fft_len: 400