Reducing Streaming ASR Model Delay with Self Alignment

05/06/2021
by   Jaeyoung Kim, et al.
4

Reducing prediction delay for streaming end-to-end ASR models with minimal performance regression is a challenging problem. Constrained alignment is a well-known existing approach that penalizes predicted word boundaries using external low-latency acoustic models. On the contrary, recently proposed FastEmit is a sequence-level delay regularization scheme encouraging vocabulary tokens over blanks without any reference alignments. Although all these schemes are successful in reducing delay, ASR word error rate (WER) often severely degrades after applying these delay constraining schemes. In this paper, we propose a novel delay constraining method, named self alignment. Self alignment does not require external alignment models. Instead, it utilizes Viterbi forced-alignments from the trained model to find the lower latency alignment direction. From LibriSpeech evaluation, self alignment outperformed existing schemes: 25 at the similar word error rate. For Voice Search evaluation,12 reductions were achieved compared to FastEmit and constrained alignment with more than 2

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset