Speech Tokenizer is Key to Consistent Representation

TL;DR

This paper presents a novel speech tokenizer that encodes both linguistic and acoustic features, significantly improving speech representation fidelity across various applications without extra training.

Contribution

It introduces an advanced speech tokenizer that captures both semantic and acoustic information, addressing limitations of previous RVQ-based methods.

Findings

Enhanced speech coding quality

Improved emotion recognition accuracy

Versatile application across speech tasks

Abstract

Speech tokenization is crucial in digital speech processing, converting continuous speech signals into discrete units for various computational tasks. This paper introduces a novel speech tokenizer with broad applicability across downstream tasks. While recent advances in residual vector quantization (RVQ) have incorporated semantic elements, they often neglect critical acoustic features. We propose an advanced approach that simultaneously encodes both linguistic and acoustic information, preserving prosodic and emotional content. Our method significantly enhances speech representation fidelity across diverse applications. Empirical evaluations demonstrate its effectiveness in speech coding, voice conversion, emotion recognition, and multimodal language modeling, without requiring additional training. This versatility underscores its potential as a key tool for advancing AI-driven speech processing.