Conv TasNet

This repository is based on JusperLee's repository.

Quick Start

runs on Linux

git clone https://github.com/jbnu-capstone/conv-tasnet-pytorch
cd conv-tasnet-pytorch
conda env create -f environment.yml; conda activate ctnp

python prepare_data.py
python train.py

Requirements

• python = 3.7
• tqdm
• pyyaml
• torchaudio
• soundfile
• matplotlib

---

Inference

python Separation_wav.py -mix_wav train.wav -model ./models/best.pt

• other options: -gpuid, -yaml, -save_path

Using Custom Data

Should have a structure like following:

data/
├── cv/
│   ├── mix/
│   ├── s1/
│   └── s2/
├── tr/
│   ├── mix/
│   ├── s1/
│   └── s2/
└── tt/
    ├── mix/
    ├── s1/
    └── s2/