Multiple-hypothesis RNN-T Loss for Unsupervised Fine-tuning and Self-training of Neural Transducer.

Cong-Thanh Do, Mohan Li, Rama Doddipatla

Browse the full Interspeech paper archive.