ART: A Faster Way to Train Speech Recognition Models

How residual learning and CTC alignment can accelerate RNN-Transducer training.

Continue reading on Medium »