Improving the Performance of Online Neural Transducer Models
This work addresses the need for efficient, real-time speech recognition in applications like voice search, representing an incremental improvement over existing methods.
The paper tackled the performance gap between online neural transducer models and non-streaming models like LAS for streaming voice search, achieving matching performance through improvements in attention window, initialization from LAS, and stronger language models.
Having a sequence-to-sequence model which can operate in an online fashion is important for streaming applications such as Voice Search. Neural transducer is a streaming sequence-to-sequence model, but has shown a significant degradation in performance compared to non-streaming models such as Listen, Attend and Spell (LAS). In this paper, we present various improvements to NT. Specifically, we look at increasing the window over which NT computes attention, mainly by looking backwards in time so the model still remains online. In addition, we explore initializing a NT model from a LAS-trained model so that it is guided with a better alignment. Finally, we explore including stronger language models such as using wordpiece models, and applying an external LM during the beam search. On a Voice Search task, we find with these improvements we can get NT to match the performance of LAS.