README.md

conv-tasnet-visualization

A implementation of Conv-TasNet visualization.

conv-tasnet-visualization use Asteroid model.

The images of visualization are produced with JorisCos/ConvTasNet_Libri2Mix_sepclean_8k.

Usage

python main.py \
  --model_path path/to/model \
  --mixture_path path/to/mixture_audio \
  --spk1_path path/to/spk1_audio \
  --spk2_path path/to/spk2_audio

• --model_path: Conv-TasNet model path.
• --mixture_path: mixture audio file path.
• --spk1_path: speaker 1 audio file path.
• --spk2_path: speaker 2 audio file path.

Visualization

The sample audio isn't provided because it is generated from WSJ0.

Waveform

Encoder and decoder basis functions

   

Encoder representation

Source masks