Enhance audio generation controllability through representation similarity regularization

TL;DR

This paper introduces a regularization technique that aligns audio and text representations during training, significantly improving controllability and quality in audio and music generation tasks.

Contribution

It proposes a novel representation similarity regularization method during training, especially in classifier-free guidance, to enhance alignment between audio and text representations.

Findings

Improved objective metrics for audio and music generation.

Enhanced human perception of generated audio.

Better control over audio generation quality.

Abstract

This paper presents an innovative approach to enhance control over audio generation by emphasizing the alignment between audio and text representations during model training. In the context of language model-based audio generation, the model leverages input from both textual and audio token representations to predict subsequent audio tokens. However, the current configuration lacks explicit regularization to ensure the alignment between the chosen text representation and the language model's predictions. Our proposal involves the incorporation of audio and text representation regularization, particularly during the classifier-free guidance (CFG) phase, where the text condition is excluded from cross attention during language model training. The aim of this proposed representation regularization is to minimize discrepancies in audio and text similarity compared to other samples within the same training batch. Experimental results on both music and audio generation tasks demonstrate that our proposed methods lead to improvements in objective metrics for both audio and music generation, as well as an enhancement in the human perception for audio generation.