A Conformer Based Acoustic Model for Robust Automatic Speech Recognition

TL;DR

This paper introduces a Conformer-based acoustic model for robust speech recognition, demonstrating significant improvements in accuracy, model size, and training efficiency on the CHiME-4 dataset.

Contribution

It replaces the recurrent network in WRBN with a Conformer encoder, achieving better performance and efficiency in speech recognition tasks.

Findings

Achieves 6.25% WER on CHiME-4, outperforming WRBN by 8.4% relative.

Model size is reduced by 18.3%, and training time is cut by 79.6%.

Uses convolution-augmented attention for improved acoustic modeling.

Abstract

This study addresses robust automatic speech recognition (ASR) by introducing a Conformer-based acoustic model. The proposed model builds on the wide residual bi-directional long short-term memory network (WRBN) with utterance-wise dropout and iterative speaker adaptation, but employs a Conformer encoder instead of the recurrent network. The Conformer encoder uses a convolution-augmented attention mechanism for acoustic modeling. The proposed system is evaluated on the monaural ASR task of the CHiME-4 corpus. Coupled with utterance-wise normalization and speaker adaptation, our model achieves $6.25\%$ word error rate, which outperforms WRBN by $8.4\%$ relatively. In addition, the proposed Conformer-based model is $18.3\%$ smaller in model size and reduces total training time by $79.6\%$.