以软体为基础建构语音增强系统使用者介面Developmentofa.pdf

The 2017 Conference on Computational Linguistics and Speech Processing ROCLING 2017, pp. 323-331  The Association for Computational Linguistics and Chinese Language Processing 以軟體為基礎建構語音增強系統使用者介面 Development of a software-based User-Interface of Speech Enhancement System 王韜維 Tao-Wei Wang 曹昱 Yu Tsao 中央研究院資訊創新科技研究中心 Research Center for Information Technology Innovation Academia Sinica 賴穎暉 Ying-Hui Lai 國立陽明大學生物醫學工程學系 Department of BioMedical Engineering National Yang-Ming University 吳家隆 Chia-Lung Wu 許祥平Hsiang-Ping Hsu 法務部調查局 Investigation Bureau, Ministry of Justice 摘要 本研究的目的在發展以軟體為基準的語音增強系統使用者介面，提供使用者一個快 速且便於操作的輔助工具。此使用者平台包含傳統的方法和基於機器學習發展的語音增 強演算法，使用者可以針對不同的噪音類型選擇演算法。處理後的語音除了可從介面上 取得語音波形圖與聲譜圖的結果，還可播放與儲存處理後的語音。本研究選用TMHINT 混車噪音與嬰兒哭聲做為驗證的測試語料。從結果與PESQ 顯示，傳統的方式可有效的 降低車噪音，但無法有效的降低嬰兒哭聲；而機器學習方式除了可降低車噪音，也可以 有效的降低嬰兒噪音。 Abstract The topic is to develop a user interface of speech enhancement in this study. This system includes of typical and based-on machine learning algorithm and provides a convenient and user-friendly interface. User can obtain waveform and spectrogram of enhancement speech 323 and play and restore the enhancement speech in this interface. TMHINT database with car noise or baby crying is used to test this noise reduction system. The results show that typical methods are only capable to reduce car noise, but the methods based-on machine learning could reduce both of these two noise. 關鍵詞：語音增強、NMF 、DDAE 、機