CleanCodec: Perceptually Guided Speech Tokenization

Arxiv CS.SD (Sound/Audio ML)

CleanCodec achieves efficient and robust speech tokenization by using perceptually guided encoding to balance reconstruction quality with token efficiency. The codec shows strong performance on downstream speech tasks.

speech-codec

Arxiv CS.SD (Sound/Audio ML)·Eugene KwekFeng LiuRui ZhangWenpeng Yin

7 days ago