This repository is the official code for the paper:
Unsupervised Speech Enhancement using Dynamical Variational Auto-Encoders, TASLP, 2022
Xiaoyu Bie, Simon Leglaive, Xavier Alameda-Pineda, Laurent Girin
[Paper]
For HumanEva-I and Human3.6M, we follow the instructions in gsps, then put all data into data directory:
- WSJ-QUT (paid): it comes from the the subset of original WSJ dataset, which was used in the CHiME-3 Challenge, corrupted by the QUT noise
- VoiceBank-DEMAND (free): download data from their website, downsample to 16kHz and select p226 & p287 as validation dataset (following MetricGanU)
If you find this code useful, please star the project and consider citing:
@article{bie2022unsupervised,
title={Unsupervised speech enhancement using dynamical variational autoencoders},
author={Bie, Xiaoyu and Leglaive, Simon and Alameda-Pineda, Xavier and Girin, Laurent},
journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing},
volume={30},
pages={2993--3007},
year={2022},
publisher={IEEE}
}