Conference proceeding
Improving The Latency And Quality Of Cascaded Encoders
ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Vol.2022-, pp.8112-8116
05/23/2022
DOI: 10.1109/ICASSP43922.2022.9747879
Abstract
In this paper, we explore reducing computational latency of the 2-pass cascaded encoder model [1]. Specifically, we experiment with reducing the size of the causal 1st-pass and adding capacity to the non-causal 2nd-pass, such that the overall latency can be reduced without loss of quality. In addition, we explore using a confidence model for deciding to stop 2nd-pass recognition if we are confident in the 1st-pass hypothesis. Overall, we are able to reduce latency by a factor of 1.7X, compared to the baseline cascaded encoder from [1]. Secondly, with the added capacity in the non-causal 2nd-pass, we find that we can improve WER by up to 7% relative using wav2vec and minimum word-error-rate (MWER) training.
Details
- Title: Subtitle
- Improving The Latency And Quality Of Cascaded Encoders
- Creators
- Tara N. Sainath - GoogleYanzhang He - GoogleArun Narayanan - GoogleRami Botros - GoogleWeiran Wang - GoogleDavid Qiu - GoogleChung-Cheng Chiu - Google LLC,USARohit Prabhavalkar - GoogleAlexander Gruenstein - GoogleAnmol Gulati - GoogleBo Li - GoogleDavid Rybach - GoogleEmmanuel Guzman - GoogleIan McGraw - GoogleJames Qin - GoogleKrzysztof Choromanski - GoogleQiao Liang - GoogleRobert David - GoogleRuoming Pang - GoogleShuo-Yiin Chang - GoogleTrevor Strohman - GoogleW. Ronny Huang - Google LLC,USAWei Han - GoogleYonghui Wu - GoogleYu Zhang - Google
- Resource Type
- Conference proceeding
- Publication Details
- ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Vol.2022-, pp.8112-8116
- Publisher
- IEEE
- DOI
- 10.1109/ICASSP43922.2022.9747879
- ISSN
- 1520-6149
- eISSN
- 2379-190X
- Language
- English
- Date published
- 05/23/2022
- Academic Unit
- Computer Science
- Record Identifier
- 9984696710602771
Metrics
1 Record Views