A PyTorch implementation of "TasNet: Time-domain Audio Separation Network for Real-time, single-channel speech separation", published in ICASSP2018, by Yi Luo and Nima Mesgarani.
- PyTorch 0.4.1+
- Python3 (Recommend Anaconda)
pip install -r requirements.txt- If you need to convert wjs0 to wav format and generate mixture files,
cd tools; make
See egs/wsj0/run.sh for example and modify related wsj0 data path.
- Stage 0: Convert sphere format to wav format and generate mixture (optional)
- Stage 1: Generating json files including wav path and duration
- Stage 2: Training
- Stage 3: Evaluate separation performance
- Stage 4: Separate speech using TasNet
| Method | Causal | SDRi | SI-SNRi | Config |
|---|---|---|---|---|
| TasNet-BLSTM (Paper) | No | 11.1 | 10.8 | |
| TasNet-BLSTM (Mine) | No | 11.84 | 11.54 | L40 N500 hidden500 layer4 lr1e-3 epoch100 bs10 |
| TasNet-BLSTM (Mine) | No | 11.77 | 11.46 | Above + L2 1e-4 |
| TasNet-BLSTM (Mine) | No | 13.07 | 12.78 | Above + L2 1e-5 |
- Layer normlization described in paper
- LSTM skip connection
- Curriculum learning