Non-native speech recognition using characteristic speech features with respect to nationality

カテゴリ: 部門大会

論文No: SS3-11

グループ名: 【C】平成29年電気学会電子・情報・システム部門大会講演論文集

発行日: 2017/09/06

タイトル(英語): Non-native speech recognition using characteristic speech features with respect to nationality

著者名: Radzikowski Kacper(早稲田大学),Wang Le(早稲田大学),吉江 修(早稲田大学)

著者名(英語): Kacper Radzikowski|Le Wang|Osamu Yoshie

キーワード: 非母語話者|多重クラスタリング|感度解析|因子分析|音声特徴学習|non-native speakers|multidimensional clustering|sensitivity analysis|factor analysis|speech featureslearning

要約(日本語): In this paper we would like to propose a method for designing a speech recognition system, targeted for non-native speakers of some particular language.The method is based on detecting and recognizing the most common speech features, amongst certain nationality speaking the different language, than its mother tongue.Having performed the extraction of most typical features from the non-native speech database, the system uses multidimensional clustering, factor analysis and sensitivity analysis to create clusters related to age, gender, speaking rate etc., and containing models used in particular cases for more robust non-native speech recognition.In this research, we used the database containing recordings of Japanese people reading sentences in English.

PDFファイルサイズ: 158 Kバイト