請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20126
標題: | 基於高斯混合模型與深度學習類神經網路之馬可夫隱藏模型在語音評分表現之比較 A Comparative Study of Speech Assessment Using GMM-HMM and DNN-HMM |
作者: | Shih-Chan Kuo 郭世展 |
指導教授: | 張智星 |
關鍵字: | 電腦輔助發音訓練,深度學習,語言評分, CAPT (Computer Assisted Pronunciation),DNN (Deep Neural Network),speech assessment, |
出版年 : | 2017 |
學位: | 碩士 |
摘要: | 電腦輔助發音訓練系統(Computer Assisted Pronuncitation Training, CAPT)為藉由電腦之語音評分系統來對使用者發音進行評分之系統,旨在提供第二外語學習者能藉由此系統針對發音上之錯誤提供好壞的回饋,達到相較真人語言教師的學習環境能有較方便快速與精準之優勢。
另外,深度學習類神經網路模型(Deep Neural Network, DNN)近年來在語音辨識上的發揚光大,為語音辨識等領域之表現提供相當大幅度的進步,而語音評分系統又與語音辨識之聲學模型息息相關,本實驗將會觀察比較使用傳統之高斯混合模型-隱藏式馬可夫模型(GMM-HMM)為語音模型之語音評分系統,與使用深度學習模型之馬可夫隨機模型(DNN-HMM)之系統在不同英語語音評分實驗下之表現。本論文將進行兩實驗,來測試DNN-HMM模型在電腦輔助發音訓練上為傳統GMM-GMM模型提供之改進。 CAPT (Computer Assisted Pronunciation Training) is a system that compute scores of pronunciation by computers which provide quality of pronunciations for L2 learners, and is also a more convient and more consistent solution compared to human teachers. DNN (Deep Neural Network), as a state-of-the-art solution for speech recognition, has been proven successfully improving the performance of speech recognition in recent years, and since the performance of speech assessment is highly correlated to acoustic model of speech recognition, we will do a comparative experiment to compare the performance between conventional GMM-HMM model and DNN-HMM model under two different tasks. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20126 |
DOI: | 10.6342/NTU201704470 |
全文授權: | 未授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-106-1.pdf 目前未授權公開取用 | 3.62 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。