Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy

Takafumi Moriya, Tomohiro Tanaka, Takahiro Shinozaki, Shinji Watanabe, Kevin Duh

研究成果: Conference contribution

12 引用 (Scopus)

抄録

When building a state-of-the-art speech recognition system, the laborious effort required by human experts in tuning numerous parameters remains a prominent obstacle. The goal of this paper is to automate the process. We propose to tune DNN-HMM based large vocabulary speech recognition systems using the covariance matrix adaptation evolution strategy (CMA-ES) with a multi-objective Pareto optimization. This optimizes systems to achieve both high-accuracy and compact model size. An additional advantage of our approach is that it is efficiently parallelizable and easily adapted to cloud computing services. We performed experiments on the Corpus of Spontaneous Japanese (CSJ) using the TSUBAME 2.5 supercomputer. Compared with a strong manually tuned configuration borrowed from a similar system, our approach automatically discovered systems with lower WER by 0.48%, and systems with 59% smaller model size while keeping WER constant. The optimized training script is released in the Kaldi speech recognition toolkit as the first publicly available recipe for Japanese large vocabulary speech recognition.

元の言語English
ホスト出版物のタイトル2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings
出版者Institute of Electrical and Electronics Engineers Inc.
ページ610-616
ページ数7
ISBN(電子版)9781479972913
DOI
出版物ステータスPublished - 2016 2 10
外部発表Yes
イベントIEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Scottsdale, United States
継続期間: 2015 12 132015 12 17

Other

OtherIEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015
United States
Scottsdale
期間15/12/1315/12/17

Fingerprint

Speech recognition
Automation
Supercomputers
Cloud computing
Covariance matrix
Tuning
Experiments

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Computer Vision and Pattern Recognition

これを引用

Moriya, T., Tanaka, T., Shinozaki, T., Watanabe, S., & Duh, K. (2016). Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy. : 2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings (pp. 610-616). [7404852] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ASRU.2015.7404852

Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy. / Moriya, Takafumi; Tanaka, Tomohiro; Shinozaki, Takahiro; Watanabe, Shinji; Duh, Kevin.

2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2016. p. 610-616 7404852.

研究成果: Conference contribution

Moriya, T, Tanaka, T, Shinozaki, T, Watanabe, S & Duh, K 2016, Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy. : 2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings., 7404852, Institute of Electrical and Electronics Engineers Inc., pp. 610-616, IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015, Scottsdale, United States, 15/12/13. https://doi.org/10.1109/ASRU.2015.7404852
Moriya T, Tanaka T, Shinozaki T, Watanabe S, Duh K. Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy. : 2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc. 2016. p. 610-616. 7404852 https://doi.org/10.1109/ASRU.2015.7404852
Moriya, Takafumi ; Tanaka, Tomohiro ; Shinozaki, Takahiro ; Watanabe, Shinji ; Duh, Kevin. / Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy. 2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2016. pp. 610-616
@inproceedings{c1ad5c43c0bc40a281aef76d3fa5e326,
title = "Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy",
abstract = "When building a state-of-the-art speech recognition system, the laborious effort required by human experts in tuning numerous parameters remains a prominent obstacle. The goal of this paper is to automate the process. We propose to tune DNN-HMM based large vocabulary speech recognition systems using the covariance matrix adaptation evolution strategy (CMA-ES) with a multi-objective Pareto optimization. This optimizes systems to achieve both high-accuracy and compact model size. An additional advantage of our approach is that it is efficiently parallelizable and easily adapted to cloud computing services. We performed experiments on the Corpus of Spontaneous Japanese (CSJ) using the TSUBAME 2.5 supercomputer. Compared with a strong manually tuned configuration borrowed from a similar system, our approach automatically discovered systems with lower WER by 0.48{\%}, and systems with 59{\%} smaller model size while keeping WER constant. The optimized training script is released in the Kaldi speech recognition toolkit as the first publicly available recipe for Japanese large vocabulary speech recognition.",
keywords = "deep neural network, evolution strategy, Japanese spontaneous speech recognition, large vocabulary speech recognition, multi-objective optimization",
author = "Takafumi Moriya and Tomohiro Tanaka and Takahiro Shinozaki and Shinji Watanabe and Kevin Duh",
year = "2016",
month = "2",
day = "10",
doi = "10.1109/ASRU.2015.7404852",
language = "English",
pages = "610--616",
booktitle = "2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - Automation of system building for state-of-the-art large vocabulary speech recognition using evolution strategy

AU - Moriya, Takafumi

AU - Tanaka, Tomohiro

AU - Shinozaki, Takahiro

AU - Watanabe, Shinji

AU - Duh, Kevin

PY - 2016/2/10

Y1 - 2016/2/10

N2 - When building a state-of-the-art speech recognition system, the laborious effort required by human experts in tuning numerous parameters remains a prominent obstacle. The goal of this paper is to automate the process. We propose to tune DNN-HMM based large vocabulary speech recognition systems using the covariance matrix adaptation evolution strategy (CMA-ES) with a multi-objective Pareto optimization. This optimizes systems to achieve both high-accuracy and compact model size. An additional advantage of our approach is that it is efficiently parallelizable and easily adapted to cloud computing services. We performed experiments on the Corpus of Spontaneous Japanese (CSJ) using the TSUBAME 2.5 supercomputer. Compared with a strong manually tuned configuration borrowed from a similar system, our approach automatically discovered systems with lower WER by 0.48%, and systems with 59% smaller model size while keeping WER constant. The optimized training script is released in the Kaldi speech recognition toolkit as the first publicly available recipe for Japanese large vocabulary speech recognition.

AB - When building a state-of-the-art speech recognition system, the laborious effort required by human experts in tuning numerous parameters remains a prominent obstacle. The goal of this paper is to automate the process. We propose to tune DNN-HMM based large vocabulary speech recognition systems using the covariance matrix adaptation evolution strategy (CMA-ES) with a multi-objective Pareto optimization. This optimizes systems to achieve both high-accuracy and compact model size. An additional advantage of our approach is that it is efficiently parallelizable and easily adapted to cloud computing services. We performed experiments on the Corpus of Spontaneous Japanese (CSJ) using the TSUBAME 2.5 supercomputer. Compared with a strong manually tuned configuration borrowed from a similar system, our approach automatically discovered systems with lower WER by 0.48%, and systems with 59% smaller model size while keeping WER constant. The optimized training script is released in the Kaldi speech recognition toolkit as the first publicly available recipe for Japanese large vocabulary speech recognition.

KW - deep neural network

KW - evolution strategy

KW - Japanese spontaneous speech recognition

KW - large vocabulary speech recognition

KW - multi-objective optimization

UR - http://www.scopus.com/inward/record.url?scp=84964555888&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84964555888&partnerID=8YFLogxK

U2 - 10.1109/ASRU.2015.7404852

DO - 10.1109/ASRU.2015.7404852

M3 - Conference contribution

AN - SCOPUS:84964555888

SP - 610

EP - 616

BT - 2015 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2015 - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

ER -