Presentation 2008-12-09
Speech Recognition by Topic Models with Continuous/Discontinuous Topic Changes
Atsushi SAKO, Yasuo ARIKI, Tomoharu IWATA, Shinji WATANABE, Takaaki HORI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we propose topic models with continuous/discontinuous topic changes, and describe experiments using MIT Open Course Ware corpus. In a real environment, acoustic and language features vary momentarily depending on speakers, speaking styles or topic changes. To accommodate these changes, speech recognition with incremental tracking of changing environments has attracted attention. We propose a language model adaptation technique by Online Topic Model for continuous topic changes, and a technique by Topic HMM for discontinuous topic changes. The experimental results showed the improvements of Word Error Rate with these topic models. Moreover, the proposed methods outperformed the batch adaptation of language model using whole speech recognition results by tracking temporal changes of topics.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Incremental tracking of changing environments / Language model adaptation / Topic model
Paper # NLC2008-33,SP2008-88
Date of Issue

Conference Information
Committee NLC
Conference Date 2008/12/2(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Speech Recognition by Topic Models with Continuous/Discontinuous Topic Changes
Sub Title (in English)
Keyword(1) Incremental tracking of changing environments
Keyword(2) Language model adaptation
Keyword(3) Topic model
1st Author's Name Atsushi SAKO
1st Author's Affiliation Guraduate School of Science and Technology, Kobe University()
2nd Author's Name Yasuo ARIKI
2nd Author's Affiliation Guraduate School of Science and Technology, Kobe University
3rd Author's Name Tomoharu IWATA
3rd Author's Affiliation NTT Communication Science Laboratories, NTT Corporation
4th Author's Name Shinji WATANABE
4th Author's Affiliation NTT Communication Science Laboratories, NTT Corporation
5th Author's Name Takaaki HORI
5th Author's Affiliation NTT Communication Science Laboratories, NTT Corporation
Date 2008-12-09
Paper # NLC2008-33,SP2008-88
Volume (vol) vol.108
Number (no) 337
Page pp.pp.-
#Pages 6
Date of Issue