Part 1. Data(Noisy wav files) &label generation This is a script to generate the noisy training data and binary label for the training of VAD net
First, please install py-webrtcvad by： pip install webrtcvad
For detailed information, please check follwing instrucion: https://github.com/wiseman/py-webrtcvad
-
The generation of noisy data addnoise.py is used to insert mute into a wav file and add noise with desired SNR. useage: python3 ./addnoise.py --clean_file ./clean --noise_file ./noise.wav --mix_file ./mix --snr 0
-
The gerneration of corresponding label by using webrtcvad. The sensitivity of VAD is from 0 to 3. python3 ./label_generation.py --mix_file ./mix --label_file ./label --sensitivity 3 --batch_size 3
The code of training is in progress.
2022.3.9
Part 2. Training spectrogram data generation usage: bash ./run wav2stft.sh
Part 3. Train the VAD net A taining code "train2.py" was uploaded. There may some errors exist. usage： python3 train2.py --dataset wsj_noisy --save_root ./model/ --gpu 0