Speech Resynthesis from Discrete Disentangled Self-Supervised Representations

TL;DR

This paper introduces a method for speech resynthesis using self-supervised discrete representations that enable controllable synthesis, disentanglement of speech features, and an ultra-lightweight codec with high efficiency and quality.

Contribution

It presents a novel approach to speech resynthesis leveraging self-supervised discrete representations for disentangling speech features and demonstrates a lightweight codec surpassing baseline quality.

Findings

Effective disentanglement of speech content, prosody, and speaker identity.

Achieved 365 bits/sec compression with improved speech quality.

Validated through subjective human evaluation and various quantitative metrics.

Abstract

We propose using self-supervised discrete representations for the task of speech resynthesis. To generate disentangled representation, we separately extract low-bitrate representations for speech content, prosodic information, and speaker identity. This allows to synthesize speech in a controllable manner. We analyze various state-of-the-art, self-supervised representation learning methods and shed light on the advantages of each method while considering reconstruction quality and disentanglement properties. Specifically, we evaluate the F0 reconstruction, speaker identification performance (for both resynthesis and voice conversion), recordings' intelligibility, and overall quality using subjective human evaluation. Lastly, we demonstrate how these representations can be used for an ultra-lightweight speech codec. Using the obtained representations, we can get to a rate of 365 bits per second while providing better speech quality than the baseline methods. Audio samples can be found under the following link: speechbot.github.io/resynthesis.