Conference proceeding
Streaming Align-Refine for Non-autoregressive Deliberation
INTERSPEECH 2022, pp.1696-1700
Interspeech
01/01/2022
DOI: 10.21437/Interspeech.2022-10715
Abstract
We propose a streaming non-autoregressive (non-AR) decoding algorithm to deliberate the hypothesis alignment of a streaming RNN-T model. Our algorithm facilitates a simple greedy decoding procedure, and at the same time is capable of producing the decoding result at each frame with limited right context, thus enjoying both high efficiency and low latency. These advantages are achieved by converting the offline Align-Refine algorithm to be streaming-compatible, with a novel transformer decoder architecture that performs local self-attentions for both text and audio, and a time-aligned cross-attention at each layer. Furthermore, we perform discriminative training of our model with the minimum word error rate (MWER) criterion, which has not been done in the non-AR decoding literature. Experiments on voice search datasets and Librispeech show that with reasonable right context, our streaming model performs as well as the offline counterpart, and discriminative training leads to further WER gain when the first-pass model has small capacity.
Details
- Title: Subtitle
- Streaming Align-Refine for Non-autoregressive Deliberation
- Creators
- Weiran Wang - Google Inc, Mountain View, CA 94043 USAKe Hu - Google Inc, Mountain View, CA 94043 USATara N. Sainath - Google Inc, Mountain View, CA 94043 USA
- Resource Type
- Conference proceeding
- Publication Details
- INTERSPEECH 2022, pp.1696-1700
- Publisher
- Isca-Int Speech Communication Assoc
- Series
- Interspeech
- DOI
- 10.21437/Interspeech.2022-10715
- ISSN
- 2308-457X
- Number of pages
- 5
- Language
- English
- Date published
- 01/01/2022
- Academic Unit
- Computer Science
- Record Identifier
- 9984696570002771
Metrics
1 Record Views