塔米爾文語音辨識系統之設計研究

本論文針對南印度、斯里蘭卡及新加坡等地常用之塔米爾語，作語音辨識之研究，根據塔米爾語之發音方式，找出149類常用單音節，以每輪每次錄製兩個一四聲不同聲調之相同單音，錄製五輪共十次之策略，作為單音訓練之方式，建立塔米爾語資料庫中，共149類的聲紋特徵。本系統萃取梅爾頻率倒頻譜係數與線性預估倒頻譜係數作為語音之特徵，並利用隱藏式馬可夫模型及音位結構學比對，建立辨識系統。在CPU時脈2.5GHz的Intel Core 2 Quad個人電腦與Ubuntu 10.04作業系統環境下，本論文實作了3,500筆塔米爾文常用語詞辨識系統，系統之正確率可達到88.74%，平均辨識時間約在1.5秒以內。而系統之總訓練時間約為1.5小時。

Tamil language is widely used in southern India, Sri Lanka and Singapore. This paper investigates the design and implementation strategies for a Tamil speech recognition system. It utilizes the speech features of the 149 common Tamil mono-syllables as the major training and recognition methodology. A training database of 10 utterances per mono-syllable is established by applying Tamil pronunciation rules. These 10 utterances are collected through reading five rounds of the same mono-syllables twice with different tones. The first pronounced pattern has high pitch of tone one, while the second one has falling pitch of tone four. Mel-frequency cepstral coefficients, linear predicted cepstral coefficients, and hidden Markov model are used as the two feature models and the syllable recognition model respectively. The recognized syllable strings are then refined by phonotactical rules to obtain the optimal result. Under the Intel Core 2 Quad 2.5 GHz personal computer and Ubuntu 10.04 operating system environment, a correct phrase recognition rate of 88.74% can be reached for a 3,500 Tamil phrase database. The average computation time for the system is within 1.5 seconds, and the training time for the systems is about one and half hours.