SoundStream: An End-to-End Neural Audio Codec

This repository is an implementation of the article with same name.

The RVQ (stands for Residual Vector Quantizer) relies on lucidrains' repository.

I built this implementation to serve my needs and some features are missing from the original article.

Missing pieces

• Denoising: this implementation is not built to denoise, so there is no conditioning signal nor Feature-wise Linear Modulation blocks.
• Bitrate scalability: for now, quantizer dropout has not been implemented.

Citations

@misc{zeghidour2021soundstream,
    title   = {SoundStream: An End-to-End Neural Audio Codec},
    author  = {Neil Zeghidour and Alejandro Luebs and Ahmed Omran and Jan Skoglund and Marco Tagliasacchi},
    year    = {2021},
    eprint  = {2107.03312},
    archivePrefix = {arXiv},
    primaryClass = {cs.SD}
}