Multi-channel Opus compression for far-field automatic speech recognition with a fixed bitrate budget

TL;DR

This paper introduces a novel multi-channel Opus compression method that reduces bandwidth while preserving spatial information, improving far-field ASR accuracy with minimal bitrate increase.

Contribution

It proposes a modified joint channel coding and spatial decorrelation transform to enhance multi-channel audio compression for ASR.

Findings

Achieves 37.5% bitrate reduction at fixed ASR performance.

Reduces word error rate by 5.1% relative at the same bitrate.

Maintains spatial information better than standard codecs.

Abstract

Automatic speech recognition (ASR) in the cloud allows the use of larger models and more powerful multi-channel signal processing front-ends compared to on-device processing. However, it also adds an inherent latency due to the transmission of the audio signal, especially when transmitting multiple channels of a microphone array. One way to reduce the network bandwidth requirements is client-side compression with a lossy codec such as Opus. However, this compression can have a detrimental effect especially on multi-channel ASR front-ends, due to the distortion and loss of spatial information introduced by the codec. In this publication, we propose an improved approach for the compression of microphone array signals based on Opus, using a modified joint channel coding approach and additionally introducing a multi-channel spatial decorrelating transform to reduce redundancy in the transmission. We illustrate the effect of the proposed approach on the spatial information retained in multi-channel signals after compression, and evaluate the performance on far-field ASR with a multi-channel beamforming front-end. We demonstrate that our approach can lead to a 37.5 % bitrate reduction or a 5.1 % relative word error rate reduction for a fixed bitrate budget in a seven channel setup.