Benchmarking Prosody Encoding in Discrete Speech Tokens

TL;DR

This paper evaluates how well discrete speech tokens derived from self-supervised models encode prosodic features, providing insights for designing better tokenization methods in speech language models.

Contribution

It offers a comprehensive analysis of prosodic encoding in discrete speech tokens, highlighting factors affecting their sensitivity to prosody and guiding improved token design.

Findings

Discrete tokens vary in prosodic sensitivity based on clustering choices.

Certain SSL models better capture prosodic features.

Guidelines for designing discrete tokens with improved prosodic encoding.

Abstract

Recently, discrete tokens derived from self-supervised learning (SSL) models via k-means clustering have been actively studied as pseudo-text in speech language models and as efficient intermediate representations for various tasks. However, these discrete tokens are typically learned in advance, separately from the training of language models or downstream tasks. As a result, choices related to discretization, such as the SSL model used or the number of clusters, must be made heuristically. In particular, speech language models are expected to understand and generate responses that reflect not only the semantic content but also prosodic features. Yet, there has been limited research on the ability of discrete tokens to capture prosodic information. To address this gap, this study conducts a comprehensive analysis focusing on prosodic encoding based on their sensitivity to the artificially modified prosody, aiming to provide practical guidelines for designing discrete tokens.