LSCodec: Low-Bitrate and Speaker-Decoupled Discrete Speech Codec

TL;DR

LSCodec is a novel speech codec that achieves low bitrate and effective speaker decoupling through a multi-stage unsupervised training process, enabling high-quality speech synthesis and voice conversion.

Contribution

It introduces a low-bitrate, speaker-decoupled discrete speech codec with a unique training framework and a single codebook, advancing speech generation and voice conversion capabilities.

Findings

Superior intelligibility and audio quality with fewer codes.

Effective speaker disentanglement demonstrated in experiments.

Training framework's effectiveness verified through ablation studies.

Abstract

Although discrete speech tokens have exhibited strong potential for language model-based speech generation, their high bitrates and redundant timbre information restrict the development of such models. In this work, we propose LSCodec, a discrete speech codec that has both low bitrate and speaker decoupling ability. LSCodec adopts a multi-stage unsupervised training framework with a speaker perturbation technique. A continuous information bottleneck is first established, followed by vector quantization that produces a discrete speaker-decoupled space. A discrete token vocoder finally refines acoustic details from LSCodec. By reconstruction evaluations, LSCodec demonstrates superior intelligibility and audio quality with only a single codebook and smaller vocabulary size than baselines. Voice conversion and speaker probing experiments prove the excellent speaker disentanglement of LSCodec, and ablation study verifies the effectiveness of the proposed training framework.