Search error risk minimization in viterbi beam search for speech recognition

Takaaki Hori, Shinji Watanabe, Atsushi Nakamura

研究成果: Conference contribution

2 引用 (Scopus)

抄録

This paper proposes a method to optimize Viterbi beam search based on search error risk minimization in large vocabulary continuous speech recognition (LVCSR). Most speech recognizers employ beam search to speed up the decoding process, in which unpromising partial hypotheses are pruned during decoding. However, the pruning step involves the risk of missing the best complete hypothesis by discarding a partial hypothesis that might grow into the best. Missing the best hypothesis is called search error. Our purpose is to reduce search error by optimizing the pruning step. While conventional methods use heuristic criteria to prune each hypothesis based on its score, rank, and so on, our proposed method introduces a pruning function that makes a more precise decision using the rich features extracted from each hypothesis. The parameters of the function can be estimated efficiently to minimize the search error risk using recognition lattices at the training step. We implemented the new method in a WFST-based decoder and achieved a significant reduction of search errors in a 200K-word LVCSR task.

元の言語English
ホスト出版物のタイトル2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings
ページ4934-4937
ページ数4
DOI
出版物ステータスPublished - 2010
外部発表Yes
イベント2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Dallas, TX
継続期間: 2010 3 142010 3 19

Other

Other2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010
Dallas, TX
期間10/3/1410/3/19

Fingerprint

Speech recognition
Continuous speech recognition
Decoding

ASJC Scopus subject areas

  • Signal Processing
  • Software
  • Electrical and Electronic Engineering

これを引用

Hori, T., Watanabe, S., & Nakamura, A. (2010). Search error risk minimization in viterbi beam search for speech recognition. : 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings (pp. 4934-4937). [5495099] https://doi.org/10.1109/ICASSP.2010.5495099

Search error risk minimization in viterbi beam search for speech recognition. / Hori, Takaaki; Watanabe, Shinji; Nakamura, Atsushi.

2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings. 2010. p. 4934-4937 5495099.

研究成果: Conference contribution

Hori, T, Watanabe, S & Nakamura, A 2010, Search error risk minimization in viterbi beam search for speech recognition. : 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings., 5495099, pp. 4934-4937, 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010, Dallas, TX, 10/3/14. https://doi.org/10.1109/ICASSP.2010.5495099
Hori T, Watanabe S, Nakamura A. Search error risk minimization in viterbi beam search for speech recognition. : 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings. 2010. p. 4934-4937. 5495099 https://doi.org/10.1109/ICASSP.2010.5495099
Hori, Takaaki ; Watanabe, Shinji ; Nakamura, Atsushi. / Search error risk minimization in viterbi beam search for speech recognition. 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings. 2010. pp. 4934-4937
@inproceedings{4316dffac8414bf1bdc4e0a48847aefb,
title = "Search error risk minimization in viterbi beam search for speech recognition",
abstract = "This paper proposes a method to optimize Viterbi beam search based on search error risk minimization in large vocabulary continuous speech recognition (LVCSR). Most speech recognizers employ beam search to speed up the decoding process, in which unpromising partial hypotheses are pruned during decoding. However, the pruning step involves the risk of missing the best complete hypothesis by discarding a partial hypothesis that might grow into the best. Missing the best hypothesis is called search error. Our purpose is to reduce search error by optimizing the pruning step. While conventional methods use heuristic criteria to prune each hypothesis based on its score, rank, and so on, our proposed method introduces a pruning function that makes a more precise decision using the rich features extracted from each hypothesis. The parameters of the function can be estimated efficiently to minimize the search error risk using recognition lattices at the training step. We implemented the new method in a WFST-based decoder and achieved a significant reduction of search errors in a 200K-word LVCSR task.",
keywords = "Pruning, Search error, Viterbi beam search, WFST",
author = "Takaaki Hori and Shinji Watanabe and Atsushi Nakamura",
year = "2010",
doi = "10.1109/ICASSP.2010.5495099",
language = "English",
isbn = "9781424442966",
pages = "4934--4937",
booktitle = "2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings",

}

TY - GEN

T1 - Search error risk minimization in viterbi beam search for speech recognition

AU - Hori, Takaaki

AU - Watanabe, Shinji

AU - Nakamura, Atsushi

PY - 2010

Y1 - 2010

N2 - This paper proposes a method to optimize Viterbi beam search based on search error risk minimization in large vocabulary continuous speech recognition (LVCSR). Most speech recognizers employ beam search to speed up the decoding process, in which unpromising partial hypotheses are pruned during decoding. However, the pruning step involves the risk of missing the best complete hypothesis by discarding a partial hypothesis that might grow into the best. Missing the best hypothesis is called search error. Our purpose is to reduce search error by optimizing the pruning step. While conventional methods use heuristic criteria to prune each hypothesis based on its score, rank, and so on, our proposed method introduces a pruning function that makes a more precise decision using the rich features extracted from each hypothesis. The parameters of the function can be estimated efficiently to minimize the search error risk using recognition lattices at the training step. We implemented the new method in a WFST-based decoder and achieved a significant reduction of search errors in a 200K-word LVCSR task.

AB - This paper proposes a method to optimize Viterbi beam search based on search error risk minimization in large vocabulary continuous speech recognition (LVCSR). Most speech recognizers employ beam search to speed up the decoding process, in which unpromising partial hypotheses are pruned during decoding. However, the pruning step involves the risk of missing the best complete hypothesis by discarding a partial hypothesis that might grow into the best. Missing the best hypothesis is called search error. Our purpose is to reduce search error by optimizing the pruning step. While conventional methods use heuristic criteria to prune each hypothesis based on its score, rank, and so on, our proposed method introduces a pruning function that makes a more precise decision using the rich features extracted from each hypothesis. The parameters of the function can be estimated efficiently to minimize the search error risk using recognition lattices at the training step. We implemented the new method in a WFST-based decoder and achieved a significant reduction of search errors in a 200K-word LVCSR task.

KW - Pruning

KW - Search error

KW - Viterbi beam search

KW - WFST

UR - http://www.scopus.com/inward/record.url?scp=78049370808&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=78049370808&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2010.5495099

DO - 10.1109/ICASSP.2010.5495099

M3 - Conference contribution

AN - SCOPUS:78049370808

SN - 9781424442966

SP - 4934

EP - 4937

BT - 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2010 - Proceedings

ER -