MLCA-AVSR: Multi-Layer Cross Attention Fusion based Audio-Visual Speech Recognition
He Wang, Pengcheng Guo, Pan Zhou, Lei Xie
TL;DR
The paper tackles robustness gaps in AVSR by enabling cross-modal representation learning through multi-layer cross-attention fusion (MLCA-AVSR). It extends prior SLCA by embedding cross-attention modules at multiple encoder depths and guiding fusion with Inter-CTC losses, using E-Branchformer encoders for both audio and visual streams. Empirical results on the MISP2022-AVSR dataset show substantial improvements over single-modality baselines and common fusion methods, ultimately achieving a new SOTA cpCER of 29.13% on Eval^sd after ROVER fusion. The approach demonstrates strong cross-modal synergy from low-level to high-level features and establishes a practical pathway to more robust AVSR in real-world noisy settings.
Abstract
While automatic speech recognition (ASR) systems degrade significantly in noisy environments, audio-visual speech recognition (AVSR) systems aim to complement the audio stream with noise-invariant visual cues and improve the system's robustness. However, current studies mainly focus on fusing the well-learned modality features, like the output of modality-specific encoders, without considering the contextual relationship during the modality feature learning. In this study, we propose a multi-layer cross-attention fusion based AVSR (MLCA-AVSR) approach that promotes representation learning of each modality by fusing them at different levels of audio/visual encoders. Experimental results on the MISP2022-AVSR Challenge dataset show the efficacy of our proposed system, achieving a concatenated minimum permutation character error rate (cpCER) of 30.57% on the Eval set and yielding up to 3.17% relative improvement compared with our previous system which ranked the second place in the challenge. Following the fusion of multiple systems, our proposed approach surpasses the first-place system, establishing a new SOTA cpCER of 29.13% on this dataset.