The DOI matches its Crossref record, with a small difference in title or year; 70 of the 71 references it cites were found in a registry or an index; none of the 16 sources its claims cite is checked for support yet.
Deep Audio-Visual Speech Recognition
Open accessAbstract
The goal of this work is to recognise phrases and sentences being spoken by a talking face, with or without the audio. Unlike previous works that have focussed on recognising a limited number of words or phrases, we tackle lip reading as an open-world problem - unconstrained natural language sentences, and in the wild videos. Our key contributions are: (1) we compare two models for lip reading, one using a CTC loss, and the other using a sequence-to-sequence loss. Both models are built on top of the transformer self-attention architecture; (2) we investigate to what extent lip reading is complementary to audio speech recognition, especially when the audio signal is noisy; (3) we introduce and publicly release a new dataset for audio-visual speech recognition, LRS2-BBC, consisting of thousands of natural sentences from British television. The models that we train surpass the performance of all previous work on a lip reading benchmark dataset by a significant margin.
Overview
CoChat Summary
## TL;DR
This paper develops Transformer-based audio-visual speech recognition models for open-world sentence-level lip reading and compares CTC vs. sequence-to-sequence training. It also introduces the LRS2-BBC dataset and shows that visual cues improve recognition, especially when audio is noisy.
## Key findings
- On LRS2-BBC lip reading, the best model is TM-seq2seq, which achieves 48.3% WER with a language model, improving by over 22% absolute over the prior 70.4% state of the art.
- On LRS3-TED lip reading, the same video-only TM-seq2seq model sets a baseline of 58.9% WER.
- For audio-visual speech recognition on clean audio, TM-CTC improves from 10.1% WER for audio-only to 8.2% WER for audio-visual; TM-seq2seq also improves from 9.7% to 8.5% on LRS2-BBC.
- With noisy audio, combining audio and video helps substantially: for LRS2-BBC, TM-CTC drops from 53.4% WER (audio-only) to 23.6% (audio-visual), and TM-seq2seq from 57.4% to 34.2%; the paper also notes improvement of up to 30% in WER under loud background noise.
## Methods
- Compared two Transformer-based architectures for character-level speech transcription: TM-seq2seq trained with cross-entropy and TM-CTC trained with CTC loss; both use separate modality encoders and optional external character-level language model decoding.
- Built and evaluated on LRS2-BBC and LRS3-TED, using video of the mouth region sampled at 25 fps and audio features computed as 321-dimensional spectral magnitudes with a 40 ms window and 10 ms hop.
- Training used a four-stage strategy: pretrain the visual front end, generate frozen visual features, train the sequence module, then fine-tune end-to-end; evaluation reports word error rate (WER), with test-time augmentation and beam search decoding.
## Limitations
- The paper states that TM-seq2seq is harder to train and takes longer: the full audio-visual model takes about 8 days to complete the curriculum on a single GeForce Titan X GPU, versus about 5 days for TM-CTC.
- It states that TM-CTC struggles with out-of-sync audio and video compared with TM-seq2seq, even after several epochs of fine-tuning; TM-seq2seq becomes robust after one epoch of calibration.
- Evident from the text: the models are evaluated primarily with WER on the reported datasets, and the dataset is built from BBC television and TED/TEDx sources, so the results may be tied to those domain settings.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 0 Not valid · 12 Not checked · 59 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Speech recognition in a noisy environment is extremely challenging, as can be seen from the significantly lower performance of the off-the-shelf Google S2T ASR baseline (over 60% performance degradation compared to clean).”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The dataset contains talking faces together with subtitles of what is said.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results in Table 2 demonstrate that the mouth movements provide important cues in speech recognition when the audio signal is noisy; and give an improvement in performance even when the audio signal is clean - for example the word error rate is reduced from 10.1% for audio only to 8.2%, when using the audio-visual TM-CTC model.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We only need to calibrate the model for one epoch for the out-ofsync effect to practically vanish.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Notably, the audio-visual models perform much better than either the video-only, or audio-only ones under the presence of loud background noise.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
46 more claims
Create a free account to see all 71 claims and to verify the 15 references still open.
- Probably not a substantial claim. 8
- It may state two separate claims as one. 15
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“[11] is the first work introducing transformers to VSR.”
From Deep learning for visual speech analysis: A survey (2024) · cites this paper as [11]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The experiments clearly showed that the transformer-based backend network performs much better than the RNN-based backend network in the sentence-level VSR task.”
From Deep learning for visual speech analysis: A survey (2024) · cites this paper as [11]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“An intuitive strategy is to leverage semantic parts (e.g., objects) pre-extracted by an off-the-shelf detector for MML [103], [104], [105], [106], [112], [204], [310].”
From Multimodal learning with transformers: A survey (2023) · cites this paper as [103]
The cited source says what the claim attributes to it.
In the citing paper
“It is built for sentence-level lip reading, a more challenging VSR problem than word-level lip reading.”
From Deep learning for visual speech analysis: A survey (2024) · cites this paper as [11]
The cited source says what the claim attributes to it.
In the citing paper
“LRS2-BBC [11] is a sentence-level audio-visual dataset with a similar data collection pipeline and data source as that LRW dataset.”
From Deep learning for visual speech analysis: A survey (2024) · cites this paper as [11]
10 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
Self-citationImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
Exploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
(2016) · Cited by 31,590
Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation
Cho, K., van Merrienboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1179 · Cited by 25,545
Research works printed but not linked: 1
- chung, 2017, Lip reading in profile, Proc Brit Mach Vis Conf
Cited by
Sign Language Transformers: Joint End-to-End Sign Language Recognition and Translation
Camgöz, N. C., Koller, O., Hadfield, S., & Bowden, R. (2020) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr42600.2020.01004 · Cited by 1,120
Multimodal learning with transformers: A survey
Xu, P., Zhu, X., & Clifton, D. A. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3275156 · Cited by 1,072
A Lip Sync Expert Is All You Need for Speech to Lip Generation In the Wild
Prajwal, K. R., Mukhopadhyay, R., Namboodiri, V. P., & Jawahar, C. V. (2020) · Proceedings of the 28th ACM International Conference on Multimedia · doi:10.1145/3394171.3413532 · Cited by 897
How, D. N. T., Hannan, M. A., Hossain Lipu, M. S., & Ker, P. J. (2019) · IEEE Access · doi:10.1109/access.2019.2942213 · Cited by 812
Metadata
- Authors
- Afouras, T., Chung, J. S., Senior, A., Vinyals, O., & Zisserman, A.
- Affiliations
- University of Oxford (GB)
- Google DeepMind (United Kingdom) (GB)
- Venue
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Published
- 2022-12-01
- Volume
- 44
- Issue
- 12
- Pages
- 8717-8727
- ISSN
- 0162-8828 (ISSN-L)
- 2160-9292
- 1939-3539
- OpenAlex
- W2890952074
- Cited by
- 801
- References
- 71
- Citations by year
- 2026: 56 · 2025: 134 · 2024: 155 · 2023: 129 · 2022: 99
- FWCI
- 40.80
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- EPSRC Programme (Seebibyte EP/M013774/1)
- EPSRC CDT
- Oxford-Google DeepMind Graduate Scholarship
- Topics
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Human-Computer Interaction › Hand Gesture Recognition Systems
- Keywords
- audio-visual speech recognition, lip reading, CTC loss, transformer self-attention, audiovisual integration, noisy audio, talking faces
- MeSH terms
- Algorithms, Humans, Lipreading, Speech, Speech Perception
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Deep Audio-Visual Speech Recognition. https://app.cochat.ai/papers/10.1109/tpami.2018.2889052
- The paper
- Afouras, T., Chung, J. S., Senior, A., Vinyals, O., & Zisserman, A. (2022). Deep Audio-Visual Speech Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence. https://doi.org/10.1109/tpami.2018.2889052
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.