The NPU-ASLP-LiAuto System Description for Visual Speech Recognition in CNVSRC 2023
He Wang, Pengcheng Guo, Wei Chen, Pan Zhou, Lei Xie
TL;DR
The paper tackles visual speech recognition in scenarios where audio is unavailable by presenting a CNVSRC 2023 VSR system that processes multi-scale lip-motion data and employs an end-to-end architecture with joint CTC/attention loss. It leverages diverse encoders (E-Branchformer, Branchformer, Conformer) and a Transformer decoder, with ROVER-based post-fusion and LM shallow fusion to boost performance. Experiments on CN-CVS and CNVSRC-Single/Multi data show CERs of 34.76% for the Single-Speaker Eval and 41.06% for the Multi-Speaker Eval after fusion, achieving top rankings across open and fixed tracks. The work demonstrates that multi-scale lip data, encoder diversity, and effective fusion strategies can significantly advance VSR performance in challenging CNVSRC benchmarks and practical scenarios without audio.
Abstract
This paper delineates the visual speech recognition (VSR) system introduced by the NPU-ASLP-LiAuto (Team 237) in the first Chinese Continuous Visual Speech Recognition Challenge (CNVSRC) 2023, engaging in the fixed and open tracks of Single-Speaker VSR Task, and the open track of Multi-Speaker VSR Task. In terms of data processing, we leverage the lip motion extractor from the baseline1 to produce multi-scale video data. Besides, various augmentation techniques are applied during training, encompassing speed perturbation, random rotation, horizontal flipping, and color transformation. The VSR model adopts an end-to-end architecture with joint CTC/attention loss, comprising a ResNet3D visual frontend, an E-Branchformer encoder, and a Transformer decoder. Experiments show that our system achieves 34.76% CER for the Single-Speaker Task and 41.06% CER for the Multi-Speaker Task after multi-system fusion, ranking first place in all three tracks we participate.