在語音訊號中，母音是最有規律的部分，因此母音的辨識即是最首要的目標。在論文裡，先將期望得到的聲音所形成的函數表示成一權重與權重函數的線性組合，並以此稱為測試函數。接著將待測聲音在離散頻譜上形成的振幅與測試函數為基礎，表示成兩者的相似程度，並且將統計模型透過測試函數與數值分析求出極大值，則此為權重。最後代入相似程度的模型，並且設定一個門檻值，大於門檻值則是期望得到的音。經由實驗測試簡單母音，得到的辨識率約80%，並且有聲音的辨識率為90%以上。

At firstly, in the speech signal, the vowels are more regular than consonants, and the most important objective is to identify it. Secondly, the function of expection sound consiss of weights and weighting functions and the function is called test function,which shows the characteristic of thefunction of expection sound.Morever, the sound of test functions and test functions compose the degree of similarity each other.The weights are through stastical model,test function,numerical analysis and optimation to find the eigenvalues of maximum value.Finally,let’s substitute weights into the degree of similarity model and set a boundary value.Larger than the boundary value is the correct sound.After we test the simple vowel，the recognition rate is about 80%,and some of these are over 90%.

[1] Lawrence R. Rabiner, A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. Proceedings of the IEEE, 77 (2), p. 257–286, February 1989.
[2] David G. Luenberger,LINEAR AND PROGRAMMING,London,1937
[3] Rabiner LR and Juang BH.An Introduction to Hidden Markov Model[J].IEEE ASSP Magazine,Jan,1986,4-16
[4] 國立台灣師範大學國音教材編輯委員會編撰“國音學 ”正中書局,台北,1992張斌,“耳鼻喉科學”,臺初版, 正中書局, 台北, 1996。
[5]林琦焜,“傅立葉分析與應用“,初版,滄海書局，台中,2010。
[6] 劉康寰,“聲波與耳”,譯自:“Waves and the ear”,台灣商務,台北,1960.
[7] 葉志強, “音叉頻譜在母音辨識上之應用”, 國立成功大學應用數學研究所碩士論文, 2002。
[8] 李曉茹, “奇異值分解在語音辨識上之應用”, 國立成功大學應用數學研究所碩士論文2003。