To provide a speech recognition device for performing word output judging with high accuracy and thereby improving speech recognition accuracy, and to provide a recognition program therefor.
The speech recognition device for recognizing speech comprises: a speech recognition means for producing a word network from input speech; a feature amount calculation means for calculating two or more preset feature amounts, based on the word network obtained by the speech recognition means, an acoustic model, and a language model; and a score calculation means for unifying the two or more feature amounts obtained by the feature amount calculation means and outputting a reliability score and a correct answer word column.
COPYRIGHT: (C)2006,JPO&NCIPI
Toru Imai
山口辰彦 他,"信頼度尺度に基づく音声認識誤りの検出および誤り訂正",電子情報通信学会技術研究報告,2003年 8月14日,Vol.103,No.263,p.7-12
松下雅彦 他,"音声入力によるWeb検索のためのキーワード認識・抽出法の改善",情報処理学会研究報告,2004年 5月28日,Vol.2004,No.57,p.43-48
小玉康広 他,"SVMを用いた複数の大語彙連続音声認識モデルの出力の混合",日本音響学会2003年春季研究発表会講演論文集-I-,2003年 3月18日,p.151-152