You can deploy this model and start earning money today!

TheNHz/
ellipsis-lrs3-trainval-autoavsr

automatic-speech-recognition

visual-speech-recognition

lrs3

webdataset

Views

No views yet

LRS3 trainval — auto_avsr mouth crops, bridged to WebDataset

The LRS3 trainval split (30 h) as [T,96,96] uint8 grayscale mouth-ROI shards with character transcripts, produced with auto_avsr's own crop geometry so results stay comparable to that project's published LRS3 numbers. Derived from TheNHz/ellipsis-lrs3-raw (itself a verified mirror of LRS3-TED, whose official distribution was discontinued).

Attribution

LRS3-TED is by Triantafyllos Afouras, Joon Son Chung… See the full description on the dataset page: https://huggingface.co/datasets/TheNHz/ellipsis-lrs3-trainval-autoavsr.