Confidence Estimation and Improvement of Speech Recognition Results Using Web Documents

2009 Fiscal Year Final Research Report

• Project/Area Number: 19700184
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Single-year Grants
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Fukuoka University
• Principal Investigator: TAKAHASHI Shinya Fukuoka University, 工学部, 助教 (40330899)
• Project Period (FY): 2007 – 2009
• Keywords: 音声認識 / 言語モデル適応 / スペクトラルクラスタリング / ウェブドキュメント / ニュース音声 / ネットワーク言語モデル / 話し言葉認識

Research Abstract

This research attempted to achieve the high accuracy speech recognition for news speech using a language model constructed from Web documents collected with keywords in speech recognition results. In addition, the method estimating confidence of speech recognition results from clustering results of collected Web documents was proposed. Moreover, the method constructing network grammar automatically was investigated for multi-word expression in oral sentences.

Research Products

• [Journal Article] Recognizer With Dynamic Alternative Path Search and It's Performance Evaluation (2010)
  • Author(s): T. Morimoto, S. Takahashi, Speech
  • Journal Title: Intelligent Automation and Computer Engineering (Lecture Notes Elec. Eng. ) Vol.52(in Printing)
  • Peer Reviewed
• [Journal Article] Automatic Construction of a FSA Language Model and Speech Recognition on it with Dynamic Alternative Path Search (2009)
  • Author(s): T. Morimoto, S. Takahashi
  • Journal Title: Proc. of the Int. Multi-Conference on Engineer and Computer Science Vol.1 Pages: 611-615
  • Peer Reviewed
• [Journal Article] Topic Specific Language Model Based on Graph Spectral Approach for Speech Recognition (2008)
  • Author(s): S. Takahashi
  • Journal Title: Trends in Intelligent Systems and Computer Engineering (Lecture Notes Elec. Eng. ) Vol.6 Pages: 497-514
  • Peer Reviewed
• [Journal Article] Automatic Closed-Caption Alignment Using Pronunciation of Speech Recognition Transcripts for Public Relations TV Program (2008)
  • Author(s): S. Takahashi, T. Morimoto, Y. Nishimoto
  • Journal Title: Proc. of the Int. Multi-Conference on Engineer and Computer Science Vol.1 Pages: 259-263
  • Peer Reviewed
• [Journal Article] News Topic Specific Language Model Based on Spectral Clustering and Web Crawling (2007)
  • Author(s): S. Takahashi, T. Morimoto, N. Tsuruta
  • Journal Title: IAENG International Journal of Computer Science Vol.34,No.2 Pages: 208-213
  • Peer Reviewed
• [Presentation] FSA言語モデルの自動構築と動的代替パスサーチによる音声認識 (2008)
  • Author(s): 高橋伸弥、森元逞
  • Organizer: 情報処理学会音声言語情報処理研究会
  • Place of Presentation: 東京工業大学
  • Year and Date: 2008-10-24
• [Presentation] 類似音素行列を用いた音声認識結果とキャプション文字列との自動対応付けに関する検討 (2008)
  • Author(s): 高橋伸弥、森元逞、西本由之
  • Organizer: 情報処理学会全国大会
  • Place of Presentation: 筑波大学
  • Year and Date: 2008-03-13
• [Presentation] Web上の類似文書にするクラスタリング結果を用いたニュース音声のトピック分割と音声認識結果の信頼度の判定 (2007)
  • Author(s): 高橋伸弥、森元逞、鶴田直之
  • Organizer: 電気関係学会九州支部第60回連合大会
  • Place of Presentation: 琉球大学
  • Year and Date: 2007-09-19