LiRA: Learning Visual Speech Representations from Audio through Self-supervision
This addresses lip-reading for applications like accessibility or surveillance, but it is incremental as it builds on existing self-supervised and cross-modal learning approaches.
The paper tackles learning visual speech representations from audio through self-supervision, proposing LiRA to predict acoustic features from unlabelled visual speech, achieving state-of-the-art performance on LRS2 with only a fraction of labelled data.
The large amount of audiovisual content being shared online today has drawn substantial attention to the prospect of audiovisual self-supervised learning. Recent works have focused on each of these modalities separately, while others have attempted to model both simultaneously in a cross-modal fashion. However, comparatively little attention has been given to leveraging one modality as a training objective to learn from the other. In this work, we propose Learning visual speech Representations from Audio via self-supervision (LiRA). Specifically, we train a ResNet+Conformer model to predict acoustic features from unlabelled visual speech. We find that this pre-trained model can be leveraged towards word-level and sentence-level lip-reading through feature extraction and fine-tuning experiments. We show that our approach significantly outperforms other self-supervised methods on the Lip Reading in the Wild (LRW) dataset and achieves state-of-the-art performance on Lip Reading Sentences 2 (LRS2) using only a fraction of the total labelled data.