Augmenting conformers with structured state-space sequence models for online speech recognition

TL;DR

This paper enhances online speech recognition by integrating structured state-space sequence models with conformers, achieving superior accuracy through a novel combination of S4 models and convolutions.

Contribution

It introduces a new method of augmenting conformers with S4 models and convolutions, demonstrating improved performance in online ASR tasks.

Findings

Best model achieves 4.01%/8.53% WER on Librispeech.

Stacking S4 with local convolution is most effective.

Augmentation outperforms extensively tuned conformer models.

Abstract

Online speech recognition, where the model only accesses context to the left, is an important and challenging use case for ASR systems. In this work, we investigate augmenting neural encoders for online ASR by incorporating structured state-space sequence models (S4), a family of models that provide a parameter-efficient way of accessing arbitrarily long left context. We performed systematic ablation studies to compare variants of S4 models and propose two novel approaches that combine them with convolutions. We found that the most effective design is to stack a small S4 using real-valued recurrent weights with a local convolution, allowing them to work complementarily. Our best model achieves WERs of 4.01%/8.53% on test sets from Librispeech, outperforming Conformers with extensively tuned convolution.