Asynchronous Articulatory Feature Recognition Using Dynamic Bayesian Networks

• Item type: SIG Technical Reports(1)
• 公開日: 2004-12-20
• タイトル: Asynchronous Articulatory Feature Recognition Using Dynamic Bayesian Networks
• タイトル / 言語: en
• タイトル: Asynchronous Articulatory Feature Recognition Using Dynamic Bayesian Networks
• 言語: eng
• 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: Centre for Speech Technology Research University of Edinburgh United Kingdom.
• 著者所属: Centre for Speech Technology Research University of Edinburgh United Kingdom.
• 著者所属: Centre for Speech Technology Research University of Edinburgh United Kingdom.
• 著者所属(英): en
• 著者所属(英): Centre for Speech Technology Research, University of Edinburgh, United Kingdom.
• 著者所属(英): en
• 著者所属(英): Centre for Speech Technology Research, University of Edinburgh, United Kingdom.
• 著者所属(英): en
• 著者所属(英): Centre for Speech Technology Research, University of Edinburgh, United Kingdom.
• 著者名: Mirjam, WESTER; Joe, FRANKEL; Simon, KING
• 著者名(英): Mirjam, Wester; Joe, Frankel; Simon, King
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: This paper builds on previous work where dynamic Bayesian networks (DBN) were proposed as a model for articulatory feature recognition. Using DBN's makes it possible to model the dependencies between features an addition to previous approaches which was found to improve feature recognition performance. The DBN results were promising giving close to the accuracy of artificial neural nets (ANNs). However the system was trained on canonical labels leading to an overly strong set of constraints on feature co-occurrence. In this study we describe and embedded training scheme which learns a set of data-driven asynchronous feature changes where supported in the data. Using a subset of the OGI Numbers corpus we describe articulatory feature recongnition experiments using both canonically- trained and asynchronous-feature DBNs. Performance using DBNs is found to exceed that of ANNs trained on an identical task giving a higher recongnition accuracy. Furthermore inter-feature dependencies result in a more structured model giving rise to fewer feature combinations in the recognition output. In addition to an empirical evaluation of this modeling approach we give a qualitative analysis investigating the asynchrony found through our data-driven method and interpreting it using linguistic knowledge.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: This paper builds on previous work where dynamic Bayesian networks (DBN) were proposed as a model for articulatory feature recognition. Using DBN's makes it possible to model the dependencies between features, an addition to previous approaches which was found to improve feature recognition performance. The DBN results were promising giving close to the accuracy of artificial neural nets (ANNs). However, the system was trained on canonical labels, leading to an overly strong set of constraints on feature co-occurrence. In this study, we describe and embedded training scheme which learns a set of data-driven asynchronous feature changes where supported in the data. Using a subset of the OGI Numbers corpus, we describe articulatory feature recongnition experiments using both canonically- trained and asynchronous-feature DBNs. Performance using DBNs is found to exceed that of ANNs trained on an identical task, giving a higher recongnition accuracy. Furthermore, inter-feature dependencies result in a more structured model, giving rise to fewer feature combinations in the recognition output. In addition to an empirical evaluation of this modeling approach, we give a qualitative analysis, investigating the asynchrony found through our data-driven method and interpreting it using linguistic knowledge.
• 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: 情報処理学会研究報告音声言語情報処理（SLP）; 巻 2004, 号 131(2004-SLP-054), p. 37-42, 発行日 2004-12-20
• 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