Classifier-based Data Selection for Lightly-Supervised Training of Acoustic Model for Lecture Transcription

• Item type: SIG Technical Reports(1)
• 公開日: 2014-07-17
• タイトル: Classifier-based Data Selection for Lightly-Supervised Training of Acoustic Model for Lecture Transcription
• タイトル / 言語: en
• タイトル: Classifier-based Data Selection for Lightly-Supervised Training of Acoustic Model for Lecture Transcription
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 講演音声認識
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: School of Informatics, Kyoto University
• 著者所属: School of Informatics, Kyoto University
• 著者所属: School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): School of Informatics, Kyoto University
• 著者名: Sheng, Li; Yuya, Akita; Tatsuya, Kawahara
• 著者名(英): Sheng, Li; Yuya, Akita; Tatsuya, Kawahara
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: The paper addresses a scheme of lightly-supervised training of acoustic model, which exploits a large amount of data with closed caption texts but not faithful transcripts. In the proposed scheme, a sequence of the closed caption text and the ASR hypothesis by the baseline system are aligned. Then, a dedicated classifier is designed and trained to select the correct one among them or reject both. It is demonstrated that the classifier can effectively filter the usable data for acoustic model training. The scheme realizes automatic training of the model with an increased amount of data. A significant improvement in the ASR accuracy is achieved from the baseline system and also in comparison with the conventional method of lightly-supervised training based on simple matching or confidence measure score.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: The paper addresses a scheme of lightly-supervised training of acoustic model, which exploits a large amount of data with closed caption texts but not faithful transcripts. In the proposed scheme, a sequence of the closed caption text and the ASR hypothesis by the baseline system are aligned. Then, a dedicated classifier is designed and trained to select the correct one among them or reject both. It is demonstrated that the classifier can effectively filter the usable data for acoustic model training. The scheme realizes automatic training of the model with an increased amount of data. A significant improvement in the ASR accuracy is achieved from the baseline system and also in comparison with the conventional method of lightly-supervised training based on simple matching or confidence measure score.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10442647
• 書誌情報: 研究報告音声言語情報処理（SLP）; 巻 2014-SLP-102, 号 4, p. 1-5, 発行日 2014-07-17
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会