Dialogue Speech Recognition by Combining Hierarchical Topic Classification and Language Model Switching

Ian R. LANE; Tatsuya KAWAHARA; Tomoko MATSUI; Satoshi NAKAMURA

doi:10.1093/ietisy/e88-d.3.446

Dialogue Speech Recognition by Combining Hierarchical Topic Classification and Language Model Switching

Ian R. LANE, Tatsuya KAWAHARA, Tomoko MATSUI, Satoshi NAKAMURA

Full Text Views

0

Cite this

Summary :

An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.

Publication: IEICE TRANSACTIONS on Information Vol.E88-D No.3 pp.446-454

Publication Date: 2005/03/01

Publicized

Online ISSN

DOI: 10.1093/ietisy/e88-d.3.446

Type of Manuscript: Special Section PAPER (Special Section on Corpus-Based Speech Technologies)

Category: Spoken Language Systems

Cite this

Copy

Ian R. LANE, Tatsuya KAWAHARA, Tomoko MATSUI, Satoshi NAKAMURA, "Dialogue Speech Recognition by Combining Hierarchical Topic Classification and Language Model Switching" in IEICE TRANSACTIONS on Information, vol. E88-D, no. 3, pp. 446-454, March 2005, doi: 10.1093/ietisy/e88-d.3.446.
Abstract: An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.
URL: https://global.ieice.org/en_transactions/information/10.1093/ietisy/e88-d.3.446/_p

Copy

@ARTICLE{e88-d_3_446,
author={Ian R. LANE, Tatsuya KAWAHARA, Tomoko MATSUI, Satoshi NAKAMURA, },
journal={IEICE TRANSACTIONS on Information},
title={Dialogue Speech Recognition by Combining Hierarchical Topic Classification and Language Model Switching},
year={2005},
volume={E88-D},
number={3},
pages={446-454},
abstract={An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.},
keywords={},
doi={10.1093/ietisy/e88-d.3.446},
ISSN={},
month={March},}

Copy

TY - JOUR
TI - Dialogue Speech Recognition by Combining Hierarchical Topic Classification and Language Model Switching
T2 - IEICE TRANSACTIONS on Information
SP - 446
EP - 454
AU - Ian R. LANE
AU - Tatsuya KAWAHARA
AU - Tomoko MATSUI
AU - Satoshi NAKAMURA
PY - 2005
DO - 10.1093/ietisy/e88-d.3.446
JO - IEICE TRANSACTIONS on Information
SN -
VL - E88-D
IS - 3
JA - IEICE TRANSACTIONS on Information
Y1 - March 2005
AB - An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.
ER -