Multilingual Audio-Visual Speech Recognition with Hybrid CTC/RNN-T Fast Conformer

TL;DR

This paper introduces a multilingual audio-visual speech recognition model using a hybrid CTC/RNN-T Fast Conformer architecture, achieving state-of-the-art results and robustness in noisy conditions across multiple languages.

Contribution

It adapts the Fast Conformer model for multimodal speech recognition with a novel hybrid CTC/RNN-T approach and expands training data with automatically transcribed multilingual datasets.

Findings

Achieves 0.8% WER on LRS3 dataset.

Reduces average WER by 11.9% on MuAViC benchmark.

Performs well in audio-only, visual-only, and combined modes.

Abstract

Humans are adept at leveraging visual cues from lip movements for recognizing speech in adverse listening conditions. Audio-Visual Speech Recognition (AVSR) models follow similar approach to achieve robust speech recognition in noisy conditions. In this work, we present a multilingual AVSR model incorporating several enhancements to improve performance and audio noise robustness. Notably, we adapt the recently proposed Fast Conformer model to process both audio and visual modalities using a novel hybrid CTC/RNN-T architecture. We increase the amount of audio-visual training data for six distinct languages, generating automatic transcriptions of unlabelled multilingual datasets (VoxCeleb2 and AVSpeech). Our proposed model achieves new state-of-the-art performance on the LRS3 dataset, reaching WER of 0.8%. On the recently introduced MuAViC benchmark, our model yields an absolute average-WER reduction of 11.9% in comparison to the original baseline. Finally, we demonstrate the ability of the proposed model to perform audio-only, visual-only, and audio-visual speech recognition at test time.