Audio Source Separation Using Variational Autoencoders and Weak Class Supervision

This repository contains supplementary material for the paper.

Demo

You can listen to some examples of our results.

Source code

• src/mix.py: Generates the SC09 mixture dataset.
• src/train.py: Training script.
• src/test.py: Testing script. It computes the BSS_EVAL metrics.
• src/experiment.py: Automates the mix-train-test scripts.
• src/demo.py: Generates the demo page.

Dataset

Speech Commands Dataset v0.02: Download

Citation (BibTeX)

If you find this repository useful, please cite our work:

@article{karamatli2019audio,
  title={Audio source separation using variational autoencoders and weak class supervision},
  author={Karamatl{\i}, Ertu{\u{g}} and Cemgil, Ali Taylan and K{\i}rb{\i}z, Serap},
  journal={IEEE Signal Processing Letters},
  volume={26},
  number={9},
  pages={1349--1353},
  year={2019},
  doi={10.1109/LSP.2019.2929440}
}