Towards Audio Codec-based Speech Separation

TL;DR

This paper introduces Codecformer, a novel speech separation method leveraging neural audio codecs to enable efficient, high-quality separation suitable for edge devices, achieving significant computational reduction while maintaining performance.

Contribution

The paper proposes a new approach to speech separation using neural audio codec embeddings, enabling efficient processing with minimal performance loss.

Findings

52x reduction in MAC at inference

Separation performance comparable to Sepformer

Effective for edge computing scenarios

Abstract

Recent improvements in neural audio codec (NAC) models have generated interest in adopting pre-trained codecs for a variety of speech processing applications to take advantage of the efficiencies gained from high compression, but these have yet been applied to the speech separation (SS) task. SS can benefit from high compression because the compute required for traditional SS models makes them impractical for many edge computing use cases. However, SS is a waveform-masking task where compression tends to introduce distortions that severely impact performance. Here we propose a novel task of Audio Codec-based SS, where SS is performed within the embedding space of a NAC, and propose a new model, Codecformer, to address this task. At inference, Codecformer achieves a 52x reduction in MAC while producing separation performance comparable to a cloud deployment of Sepformer. This method charts a new direction for performing efficient SS in practical scenarios.