UniCodec: Unified Audio Codec with Single Domain-Adaptive Codebook

TL;DR

UniCodec introduces a single, adaptive codebook for multi-domain audio compression, leveraging domain-specific strategies and self-supervised learning to outperform existing codecs across speech, music, and sound.

Contribution

It presents a novel unified audio codec with a domain-adaptive codebook and Mixture-of-Experts, enabling effective multi-domain audio processing within a single model.

Findings

Outperforms existing unified neural codecs in audio reconstruction.

Surpasses state-of-the-art domain-specific codecs in acoustic and semantic tasks.

Achieves excellent performance across speech, music, and sound domains.

Abstract

The emergence of audio language models is empowered by neural audio codecs, which establish critical mappings between continuous waveforms and discrete tokens compatible with language model paradigms. The evolutionary trends from multi-layer residual vector quantizer to single-layer quantizer are beneficial for language-autoregressive decoding. However, the capability to handle multi-domain audio signals through a single codebook remains constrained by inter-domain distribution discrepancies. In this work, we introduce UniCodec, a unified audio codec with a single codebook to support multi-domain audio data, including speech, music, and sound. To achieve this, we propose a partitioned domain-adaptive codebook method and domain Mixture-of-Experts strategy to capture the distinct characteristics of each audio domain. Furthermore, to enrich the semantic density of the codec without auxiliary modules, we propose a self-supervised mask prediction modeling approach. Comprehensive objective and subjective evaluations demonstrate that UniCodec achieves excellent audio reconstruction performance across the three audio domains, outperforming existing unified neural codecs with a single codebook, and even surpasses state-of-the-art domain-specific codecs on both acoustic and semantic representation capabilities.