Variable Bitrate Residual Vector Quantization for Audio Coding

TL;DR

This paper introduces variable bitrate residual vector quantization (VRVQ) for audio coding, enabling adaptive codebook usage per frame and improving rate-distortion performance over fixed approaches.

Contribution

It proposes a novel VRVQ method with a gradient estimation technique for non-differentiable operations, enhancing training and outperforming existing codecs.

Findings

VRVQ achieves better rate-distortion tradeoff than fixed RVQ.

The gradient estimation method improves training stability.

Enhanced performance when integrated with state-of-the-art codecs.

Abstract

Recent state-of-the-art neural audio compression models have progressively adopted residual vector quantization (RVQ). Despite this success, these models employ a fixed number of codebooks per frame, which can be suboptimal in terms of rate-distortion tradeoff, particularly in scenarios with simple input audio, such as silence. To address this limitation, we propose variable bitrate RVQ (VRVQ) for audio codecs, which allows for more efficient coding by adapting the number of codebooks used per frame. Furthermore, we propose a gradient estimation method for the non-differentiable masking operation that transforms from the importance map to the binary importance mask, improving model training via a straight-through estimator. We demonstrate that the proposed training framework achieves superior results compared to the baseline method and shows further improvement when applied to the current state-of-the-art codec.