RepCodec: A Speech Representation Codec for Speech Tokenization

TL;DR

RepCodec is a novel speech representation codec that improves semantic speech tokenization by better preserving information, leading to enhanced performance in speech understanding and generation across multiple languages.

Contribution

It introduces a new codec that learns a vector quantization codebook from speech representations, outperforming traditional clustering methods in speech tokenization.

Findings

Significantly outperforms k-means clustering in speech tasks.

Effective across various speech encoders and languages.

Enhances information retention in speech tokenization.

Abstract

With recent rapid growth of large language models (LLMs), discrete speech tokenization has played an important role for injecting speech into LLMs. However, this discretization gives rise to a loss of information, consequently impairing overall performance. To improve the performance of these discrete speech tokens, we present RepCodec, a novel speech representation codec for semantic speech tokenization. In contrast to audio codecs which reconstruct the raw audio, RepCodec learns a vector quantization codebook through reconstructing speech representations from speech encoders like HuBERT or data2vec. Together, the speech encoder, the codec encoder and the vector quantization codebook form a pipeline for converting speech waveforms into semantic tokens. The extensive experiments illustrate that RepCodec, by virtue of its enhanced information retention capacity, significantly outperforms the widely used k-means clustering approach in both speech understanding and generation. Furthermore, this superiority extends across various speech encoders and languages, affirming the robustness of RepCodec. We believe our method can facilitate large language modeling research on speech processing.