Attention-based Neural Beamforming Layers for Multi-channel Speech Recognition

TL;DR

This paper introduces a 2D Conv-Attention module combining CNNs and attention mechanisms to improve multi-channel speech recognition, demonstrating a 3.8% relative WER reduction over baseline models.

Contribution

It proposes a novel 2D Conv-Attention layer that explicitly models channel correlations, enhancing neural beamforming for speech recognition.

Findings

Achieved a 3.8% relative WER improvement over baseline neural beamformers.

Demonstrated effectiveness of combining convolutional and attention mechanisms.

Validated on an in-house multi-channel speech dataset.

Abstract

Attention-based beamformers have recently been shown to be effective for multi-channel speech recognition. However, they are less capable at capturing local information. In this work, we propose a 2D Conv-Attention module which combines convolution neural networks with attention for beamforming. We apply self- and cross-attention to explicitly model the correlations within and between the input channels. The end-to-end 2D Conv-Attention model is compared with a multi-head self-attention and superdirective-based neural beamformers. We train and evaluate on an in-house multi-channel dataset. The results show a relative improvement of 3.8% in WER by the proposed model over the baseline neural beamformer.