ニュース番組自動字幕化のための音声認識システム  [in Japanese] A Broadcast News Transcription System for Captioning  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

高齢者や聴覚障害者への放送サービスを充実させるため、音声認識を利用した放送番組の自動字幕化を検討している。本報告では、現在開発中のニュース音声認識システムの概要を述べる。アナウンサーの音声を認識するデコーダーは、bigramを用いた単語依存N-bestに基づく第1パスと、trigramによるリスコアリングの第2パスで構成される。語彙サイズを5Kから65Kまで変化させ、音素ネットワークがリニアと木構造の場合について、認識率と処理時間を調べた。さらに、ニュースの特徴を生かした時期依存言語モデルと、電子原稿を利用した認識結果の修正について述べる。Automatic captioning for TV shows is required by old ages and hearing impaired. This paper describes a broadcast news transcription system for captioning, which is under development. A decoder converting announcers' speech into texts consists of two passes: the first pass based on word-dependent N-best search with bigram and the second pass for rescoring with trigram. Recognition accuracy and processing time were examined with a linear structured or tree structured phoneme network for some vocabulary sizes from 5K to 65K. This paper also describes a time dependent language model updated with latest news and post-correction of the transcriptions by electronic draft scripts.

Automatic captioning for TV shows is required by old ages and hearing impaired. This paper describes a broadcast news transcription system for captioning, which is under development. A decoder converting announcers' speech into texts consists of two passes : the first pass based on word-dependent N-best search with bigram and the second pass for rescoring with trigram. Recognition accuracy and processing time were examined with a linear structured or tree structured phoneme network for some vocabulary sizes from 5K to 65K. This paper also describes a time dependent language model updated with latest news and post-correction of the transcriptions by electronic draft scripts.

Journal

  • 情報処理学会研究報告ヒューマンコンピュータインタラクション(HCI)

    情報処理学会研究報告ヒューマンコンピュータインタラクション(HCI) 1998(95(1998-HI-080)), 59-64, 1998-10-16

    Information Processing Society of Japan (IPSJ)

References:  11

Cited by:  13

Codes

  • NII Article ID (NAID)
    110002942443
  • NII NACSIS-CAT ID (NCID)
    AA1221543X
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    09196072
  • Data Source
    CJP  CJPref  NII-ELS  IPSJ 
Page Top