Addressing Index Collapse of Large-Codebook Speech Tokenizer with Dual-Decoding Product-Quantized Variational Auto-Encoder

TL;DR

This paper introduces PQ-VAE, a novel speech tokenizer that uses multiple codebooks with fewer codewords to prevent index collapse, enabling larger codebooks and improving speech synthesis quality.

Contribution

It proposes a product-quantized VAE with dual-decoding training to effectively utilize large codebooks and address index collapse in speech tokenization.

Findings

PQ-VAE effectively prevents index collapse in large codebooks.

The dual-decoding training strategy improves codebook perplexity and reconstruction quality.

PQ-VAE enhances speech generation quality in language-model-based TTS.

Abstract

VQ-VAE, as a mainstream approach of speech tokenizer, has been troubled by ``index collapse'', where only a small number of codewords are activated in large codebooks. This work proposes product-quantized (PQ) VAE with more codebooks but fewer codewords to address this problem and build large-codebook speech tokenizers. It encodes speech features into multiple VQ subspaces and composes them into codewords in a larger codebook. Besides, to utilize each VQ subspace well, we also enhance PQ-VAE via a dual-decoding training strategy with the encoding and quantized sequences. The experimental results demonstrate that PQ-VAE addresses ``index collapse" effectively, especially for larger codebooks. The model with the proposed training strategy further improves codebook perplexity and reconstruction quality, outperforming other multi-codebook VQ approaches. Finally, PQ-VAE demonstrates its effectiveness in language-model-based TTS, supporting higher-quality speech generation with larger codebooks.