在本論文中，我們主要針對華語的語言學習提供一套偵測發音錯誤的方法策略，並應用於電腦輔助語言學習系統中。在我們的系統中，可依序分為兩大部份。首先是句子驗證：用於禁止使用者分心(off-task)時所閱讀朗讀課文文句，以增加第二步驟的辨識率。我們將建立兩個聲學模型，其中一個為正確發音模型(null hypothesis model)，另一為相對發音模型(alternative hypothesis model)去呈現學生語句的高低相似度模型。然後，我們使用聲學模型間的似然比(likelihood ratio test)去檢驗這兩個相似度之間的差異度，藉以判斷學生的狀態是否為分心。第二個部份為音節識別：用於辨識語者所朗讀的語音文字內容，從中找出語者發音錯誤的音節並給予提示回饋。我們對於每個發音找尋與其相應的混淆音，製作成發音混淆表，並經由發音混淆表將原本的搜尋網路擴展成發音混淆搜尋網路，由語音在通過發音混淆搜尋網路所得到的語音辨識結果中，找出與正確答案不一致的音節，並告知語者發音錯誤所在。

In this thesis, we provide a strategy of error detection of pronunciation, which mainly aims at the mandarin language learning, and apply it to the computer assistant language learning system. In our system, it can be divided into two parts in sequence. First is the sentence verification, which is used to ban off-task sentences for increasing the accuracy in the second step. We establish two acoustic models, where one is null hypothesis model and the other is alternative hypothesis model, to represent to the highly-similar pronunciation of the student sentence and the low-similar one respectively. Then, we use the likelihood ratio test, which is computed between the maximum probability of a result under two different hypotheses, to verify the deviation degree between those two hypotheses and decide whether the student pronunciation is off-task or not. The second part is the syllable Identification. It is applied to recognize the content of speech read by the speaker, and find out the error pronounced syllable for the appropriate feedback to correct the pronunciation of the users. The sausage net is built with pronunciation confusion table corresponding to confusion error patterns. From the syllable identification results passing through the pronunciation confusion searching net, we find out the syllables inconsistent with the correct answer and inform where the speaker misread.

第一章 緒論 - 1 -
1.1研究動機及目的 - 1 -
1.2 研究背景及文獻探討 - 2 -
1.3研究方法 - 2 -
1.4 章節概要 - 3 -
第二章 語言學簡介與語音辨識原理 - 5 -
2.1 語言學簡介 - 5 -
2.2 語音辨識原理 - 7 -
2.2.1語音辨識基本概念 - 7 -
2.2.2 語音辨識架構 - 10 -
2.2.3 特徵擷取 - 11 -
2.2.4 隱藏式馬可夫模型 - 18 -
2.2.5 維特比搜尋 - 21 -
第三章 句子驗證(Sentence Verification) - 24 -
3.1模型的訓練 - 24 -
3.2評估方法 - 26 -
第四章 音節識別（Syllable Identification） - 29 -
4.1 音節的定義及分類 - 29 -
4.2音節識別 - 30 -
4.2.1建立發音混淆搜尋網路(Pronunciation Confusion searching net) - 30 -
4.2.2 混淆表之建立 - 31 -
第五章 實驗結果 - 35 -
5.1 實驗簡介 - 35 -
5.2 語料和聲學模型 - 35 -
5.2.1聲學模型之訓練語料與建立 - 35 -
5.3 實驗一：尋找句子驗證方法中，最好的正確發音模型與相對發音模型 - 39 -
5.4 實驗二：使用句子驗證之方法區別混淆發音與不像發音 - 46 -
5.5 實驗三：使用音節識別區分正確答案與混淆發音 - 48 -
5.6 實驗四：利用句子驗證過濾不像發音在使用音節識別偵　　　　　　測發音上的錯誤 - 50 -
第六章 系統實作 - 52 -
6.1 系統架構 - 52 -
6.2 系統介面 - 54 -
第七章 結論與未來展望 - 55 -
參考文獻 - 56 -

[1] 貝瑞‧法柏”How to learn any language”
[2] S. Wei, et al., “Automatic Mandarin Pronunciation Scoring for Native Learners with Dialect Accent”, In Proc. Interspeech 2006, Pittsburgh, Pennsylvania, 2006.
[3] H.Franco, et al., “Automatic Pronunciation Scoring for Language Instruction” , In Proc. ICASSP 97, Munich, Germany, April 21-24, 1997
[4] A. Neri, et al., “ASR-based Corrective Feedback on Pronunciation: does it really work?”, In Proc. Interspeech 2006, Pittsburgh, Pennsylvania, 2006.
[5] IPA：http://www.arts.gla.ac.uk/ipa/ipachart.html
[6] 王小川,”語音信號處理”
[7] L.R. Rabiner and B.H. Juang, Fundamentals of Speech Recognition, Prentice Hall,1993.
[8] L.R. Rabiner and B.H. Juang, “An Introduction to Hidden Markov Model,” IEEE ASSP Magazine, pp. 4-16, Jun. 1986.
[9] L.R. Rabiner, “A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition,” Proc. of the IEEE, Vol. 77, No.2, pp. 257-286, Feb. 1989.