# Audio-visual Speech Recognition on TED LRS3

Source: [Audio-visual Speech Recognition on TED LRS3](<https://www.sota2.com/research/sota/audio-visual-speech-recognition-on-ted-lrs3>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T06:08:27.006Z

Evaluation of models performing Audio-visual Speech Recognition on the TED LRS3 dataset, measured using Word Error Rate (WER).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Audio-visual Speech Recognition |
| Dataset (including any stated split) | TED LRS3 |
| Metric | WER |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 10 of 10 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [WER](<https://www.sota2.com/research/sota/audio-visual-speech-recognition-on-ted-lrs3.md?metric=WER>): Lower is better; unit: PERCENT.

## Recorded results

| Method | WER | Result date | Report paper |
| --- | --- | --- | --- |
| VGG CONFORMER | 0.009 | Not recorded | 68fa717c1e209131cea5195f |
| LP CONFORMER | 0.009 | Not recorded | 68fa717c1e209131cea5195f |
| U-HUBERT | 0.012 | Not recorded | 68fa717c1e209131cea5195f |
| AV-HUBERT | 0.013 | Not recorded | 68fa717c1e209131cea5195f |
| VIT 3D | 0.016 | Not recorded | 68fa717c1e209131cea5195f |
| RESNET CONFORMER | 0.023 | Not recorded | 68fa717c1e209131cea5195f |
| VGG TRANSFORMER | 0.03 | Not recorded | 68fa717c1e209131cea5195f |
| RNN-T | 0.045 | Not recorded | 68fa717c1e209131cea5195f |
| TM-SEQ2SEQ | 0.08 | Not recorded | 68fa717c1e209131cea5195f |
| TM-CTC | 0.123 | Not recorded | 68fa717c1e209131cea5195f |

### VGG CONFORMER

WER: 0.009. Result ID: 6991d2a33a6c3148490acfca.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718d1e209131cea51ef0.md>)

Recorded evaluation configuration: {"Visual Front-End":"VGG","Encoder":"CONFORMER","Train Data (Hours)":"100K"}. This record does not assert independent reproduction.

### LP CONFORMER

WER: 0.009. Result ID: 6991d2a33a6c3148490acfcb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

Recorded evaluation configuration: {"Visual Front-End":"LINEAR PROJECTION","Encoder":"CONFORMER","Train Data (Hours)":"100K"}. This record does not assert independent reproduction.

### U-HUBERT

WER: 0.012. Result ID: 6991d2a33a6c3148490acfc7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e0d1e209131cea3dafe.md>)

[Implementation](<https://github.com/facebookresearch/av_hubert>)

Recorded evaluation configuration: {"Visual Front-End":"RESNET","Encoder":"TRANSFORMER","Train Data (Hours)":"2K"}. This record does not assert independent reproduction.

### AV-HUBERT

WER: 0.013. Result ID: 6991d2a33a6c3148490acfc8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bff1e209131cea31879.md>)

[Implementation](<https://github.com/facebookresearch/av_hubert>)

Recorded evaluation configuration: {"Visual Front-End":"RESNET","Encoder":"TRANSFORMER","Train Data (Hours)":"2K"}. This record does not assert independent reproduction.

### VIT 3D

WER: 0.016. Result ID: 6991d2a33a6c3148490acfc9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6dc41e209131cea3bf87.md>)

Recorded evaluation configuration: {"Visual Front-End":"VIT","Encoder":"CONFORMER","Train Data (Hours)":"90K"}. This record does not assert independent reproduction.

### RESNET CONFORMER

WER: 0.023. Result ID: 6991d2a33a6c3148490acfc4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69501e209131cea2177e.md>)

[Implementation](<https://sites.google.com/view/audiovisual-speech-recognition>)

Recorded evaluation configuration: {"Visual Front-End":"CONV3D+RESNET","Encoder":"CONFORMER","Train Data (Hours)":"0.5K"}. This record does not assert independent reproduction.

### VGG TRANSFORMER

WER: 0.03. Result ID: 6991d2a33a6c3148490acfc6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c891e209131cea34896.md>)

Recorded evaluation configuration: {"Visual Front-End":"VGG","Encoder":"TRANSFORMER","Train Data (Hours)":"90K"}. This record does not assert independent reproduction.

### RNN-T

WER: 0.045. Result ID: 6991d2a33a6c3148490acfc5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66ae1e209131cea11b3b.md>)

Recorded evaluation configuration: {"Visual Front-End":"CONV3D","Encoder":"LSTM","Train Data (Hours)":"31K"}. This record does not assert independent reproduction.

### TM-SEQ2SEQ

WER: 0.08. Result ID: 6991d2a33a6c3148490acfc3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65321e209131cea08d0c.md>)

Recorded evaluation configuration: {"Visual Front-End":"RESNET","Encoder":"TRANSFORMER","Train Data (Hours)":"0.5K"}. This record does not assert independent reproduction.

### TM-CTC

WER: 0.123. Result ID: 6991d2a33a6c3148490acfc2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa717c1e209131cea5195f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65321e209131cea08d0c.md>)

Recorded evaluation configuration: {"Visual Front-End":"RESNET","Encoder":"TRANSFORMER","Train Data (Hours)":"0.5K"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
