Master’s Thesis — NLP Lip-Reading

Seq2Seq (GRU–attention), features (phoneme–viseme), TPU training; ~3% WER and 0.92 BLEU.