V3 seq2seq decoder export (TransformerDecoder, 6 layers, vocab=1000)

V3 seq2seq decoder export (TransformerDecoder, 6 layers, vocab=1000)

ONNX-exported AV-HuBERT base+vox+433h fine-tuned visual encoder (Docker, parity-passed)

ONNX-exported AV-HuBERT base visual encoder (Docker route, parity-fixed)

Add AV-HuBERT license (per redistribution requirement)

Add README with attribution and provenance

Mirror of facebookresearch AV-HuBERT large_vox_iter5 (LRS3+VoxCeleb2 pretrained)

initial commit