透過您的圖書館登入
IP:13.59.82.167

摘要


本文介紹前瞻技術中心所發展的中英文混雜關鍵詞萃取技術;以先前應用於中文自動總機系統的關鍵詞萃取技術為基礎[1],開發符合國人說話習慣的中英文混雜關鍵詞萃取技術。文章將對每個環節進行詳盡地討論和介紹,尤其在中英文混雜的部分,我們除了提出了一個中英文辨識分數的偏差補償,來克服中英文聲學模型間的差異性之外,對於關鍵詞的驗證,也提出一個正規化機制,來對結合各種不同驗證機制所形成的驗證分數進行調和,以從辨識所產生的各種結果裡,挑出其中最有可能的結果,作為最終的輸出。這項技術的開發,除了建立了我們在英文語音辨識的基礎之外,也大幅增加了語音辨識技術的實用性。

並列摘要


Base on our former experience in the ”ITRI 104 Auto Attendant System”[1] of using keyword spotting for Mandarin speech recognition, a Chinese-English mixed-lingual keyword spotting for catering the speaking style of Tawanese is presented. Detailed descriptions and discussions for developing the mixed-lingual auto-attendant system are included in this paper, especially for integrating different scoring mechanisms in the recognition phase and the verification phase for these two languages. In the recognition phase, we propose a bias-compensation method to make up the score-gap in the likelihood calculation of using Chinese and English acoustic models. To select the most probable result from the recognized hypotheses, the method for normalizing the combination scores of using different verification mechanisms in the verification phase is a1so presented.

被引用紀錄


許順翔(2009)。基於隱藏式條件隨機域聲學模型之資源受限裝置語音命令系統〔碩士論文,元智大學〕。華藝線上圖書館。https://doi.org/10.6838/YZU.2009.00304
陳弘啟(2009)。資源受限運算環境下華英混雜語音辨識系統〔碩士論文,元智大學〕。華藝線上圖書館。https://doi.org/10.6838/YZU.2009.00155

延伸閱讀