Handling uncertain observations in unsupervised topic-mixture language model adaptation

Ekapol Chuangsuwanich, Shinji Watanabe, Takaaki Hori, Tomoharu Iwata, James Glass

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

We propose an extension to the recent approaches in topic-mixture modeling such as Latent Dirichlet Allocation and Topic Tracking Model for the purpose of unsupervised adaptation in speech recognition. Instead of using the 1-best input given by the speech recognizer, the proposed model takes confusion network as an input to alleviate recognition errors. We incorporate a selection variable which helps reweight the recognition output, thus creating a more accurate latent topic estimate. Compared to adapting based on just one recognition hypothesis, the proposed model show WER improvements on two different tasks.

Original languageEnglish
Title of host publication2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings
Pages5033-5036
Number of pages4
DOIs
Publication statusPublished - 2012
Externally publishedYes
Event2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Kyoto
Duration: 2012 Mar 252012 Mar 30

Other

Other2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012
CityKyoto
Period12/3/2512/3/30

Fingerprint

Speech recognition

Keywords

  • confusion network
  • language model
  • latent topic model
  • topic tracking

ASJC Scopus subject areas

  • Signal Processing
  • Software
  • Electrical and Electronic Engineering

Cite this

Chuangsuwanich, E., Watanabe, S., Hori, T., Iwata, T., & Glass, J. (2012). Handling uncertain observations in unsupervised topic-mixture language model adaptation. In 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings (pp. 5033-5036). [6289051] https://doi.org/10.1109/ICASSP.2012.6289051

Handling uncertain observations in unsupervised topic-mixture language model adaptation. / Chuangsuwanich, Ekapol; Watanabe, Shinji; Hori, Takaaki; Iwata, Tomoharu; Glass, James.

2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. 2012. p. 5033-5036 6289051.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Chuangsuwanich, E, Watanabe, S, Hori, T, Iwata, T & Glass, J 2012, Handling uncertain observations in unsupervised topic-mixture language model adaptation. in 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings., 6289051, pp. 5033-5036, 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012, Kyoto, 12/3/25. https://doi.org/10.1109/ICASSP.2012.6289051
Chuangsuwanich E, Watanabe S, Hori T, Iwata T, Glass J. Handling uncertain observations in unsupervised topic-mixture language model adaptation. In 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. 2012. p. 5033-5036. 6289051 https://doi.org/10.1109/ICASSP.2012.6289051
Chuangsuwanich, Ekapol ; Watanabe, Shinji ; Hori, Takaaki ; Iwata, Tomoharu ; Glass, James. / Handling uncertain observations in unsupervised topic-mixture language model adaptation. 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. 2012. pp. 5033-5036
@inproceedings{5f4ab1450ac34893a8784bc1132162ca,
title = "Handling uncertain observations in unsupervised topic-mixture language model adaptation",
abstract = "We propose an extension to the recent approaches in topic-mixture modeling such as Latent Dirichlet Allocation and Topic Tracking Model for the purpose of unsupervised adaptation in speech recognition. Instead of using the 1-best input given by the speech recognizer, the proposed model takes confusion network as an input to alleviate recognition errors. We incorporate a selection variable which helps reweight the recognition output, thus creating a more accurate latent topic estimate. Compared to adapting based on just one recognition hypothesis, the proposed model show WER improvements on two different tasks.",
keywords = "confusion network, language model, latent topic model, topic tracking",
author = "Ekapol Chuangsuwanich and Shinji Watanabe and Takaaki Hori and Tomoharu Iwata and James Glass",
year = "2012",
doi = "10.1109/ICASSP.2012.6289051",
language = "English",
isbn = "9781467300469",
pages = "5033--5036",
booktitle = "2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings",

}

TY - GEN

T1 - Handling uncertain observations in unsupervised topic-mixture language model adaptation

AU - Chuangsuwanich, Ekapol

AU - Watanabe, Shinji

AU - Hori, Takaaki

AU - Iwata, Tomoharu

AU - Glass, James

PY - 2012

Y1 - 2012

N2 - We propose an extension to the recent approaches in topic-mixture modeling such as Latent Dirichlet Allocation and Topic Tracking Model for the purpose of unsupervised adaptation in speech recognition. Instead of using the 1-best input given by the speech recognizer, the proposed model takes confusion network as an input to alleviate recognition errors. We incorporate a selection variable which helps reweight the recognition output, thus creating a more accurate latent topic estimate. Compared to adapting based on just one recognition hypothesis, the proposed model show WER improvements on two different tasks.

AB - We propose an extension to the recent approaches in topic-mixture modeling such as Latent Dirichlet Allocation and Topic Tracking Model for the purpose of unsupervised adaptation in speech recognition. Instead of using the 1-best input given by the speech recognizer, the proposed model takes confusion network as an input to alleviate recognition errors. We incorporate a selection variable which helps reweight the recognition output, thus creating a more accurate latent topic estimate. Compared to adapting based on just one recognition hypothesis, the proposed model show WER improvements on two different tasks.

KW - confusion network

KW - language model

KW - latent topic model

KW - topic tracking

UR - http://www.scopus.com/inward/record.url?scp=84867589374&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84867589374&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2012.6289051

DO - 10.1109/ICASSP.2012.6289051

M3 - Conference contribution

AN - SCOPUS:84867589374

SN - 9781467300469

SP - 5033

EP - 5036

BT - 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings

ER -