Papers › Sub-word Level Lip Reading With Visual Attention

Sub-word Level Lip Reading With Visual Attention

14 Oct 2021CVPR 2022 1arXiv:2110.07603archive 2025-07-28

K R Prajwal, Triantafyllos Afouras, Andrew Zisserman

The goal of this paper is to learn strong lip reading models that can recognise speech in silent videos. Most prior works deal with the open-set visual speech recognition problem by adapting existing automatic speech recognition techniques on top of trivially pooled visual features. Instead, in this paper we focus on the unique challenges encountered in lip reading and propose tailored solutions. To this end, we make the following contributions: (1) we propose an attention-based pooling mechanism to aggregate visual speech representations; (2) we use sub-word units for lip reading for the first time and show that this allows us to better model the ambiguities of the task; (3) we propose a model for Visual Speech Detection (VSD), trained on top of the lip reading network. Following the above, we obtain state-of-the-art results on the challenging LRS2 and LRS3 benchmarks when training on public datasets, and even surpass models trained on large-scale industrial datasets by using an order of magnitude less data. Our best model achieves 22.6% word error rate on the LRS2 dataset, a performance unprecedented for lip reading models, significantly reducing the performance gap between lip reading and automatic speech recognition. Moreover, on the AVA-ActiveSpeaker benchmark, our VSD model surpasses all visual-only baselines and even outperforms several recent audio-visual methods.

PaperPDFConference PDF

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Audio-Visual Active Speaker DetectionAutomatic Speech RecognitionAutomatic Speech Recognition (ASR)Lip ReadingLipreadingSpeech RecognitionVisual Speech Recognitionspeech-recognition

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Audio-Visual Active Speaker Detection AVA-ActiveSpeaker VTP (visual only) validation mean average precision 89.2% #15 of 20 Archive leaderboard report
Lipreading LRS2 VTP (more data) Word Error Rate (WER) 22.6 #5 of 25 Archive leaderboard report
Lipreading LRS2 VTP Word Error Rate (WER) 28.9 #10 of 25 Archive leaderboard report
Lipreading LRS3-TED VTP (more data) Word Error Rate (WER) 30.7 #11 of 23 Archive leaderboard report
Lipreading LRS3-TED VTP Word Error Rate (WER) 40.6 #17 of 23 Archive leaderboard report
Visual Speech Recognition LRS2 VTP with more data Word Error Rate (WER) 22.6 #1 of 2 Archive leaderboard report
Visual Speech Recognition LRS2 VTP Word Error Rate (WER) 28.9 #2 of 2 Archive leaderboard report
Visual Speech Recognition LRS3-TED VTP with more data Word Error Rate (WER) 30.7 #2 of 3 Archive leaderboard report
Visual Speech Recognition LRS3-TED VTP Word Error Rate (WER) 40.6 #3 of 3 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Absolute Position EncodingsAdamAttentionBPEDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSoftmaxTransformerWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections