自發性中文語音基本辨認系統之建立

Abstract

在本論文中，我們建立一個自發性中文對話語音辨識基本系統架構，探討中文語音及自發性語料的特殊語音現象，如感嘆詞(particles)、不確定語音發音（uncertain sounds）、非語音聲音(paralinguistic sounds)等。我們使用中研院提供的八段雙人對話語料庫做實驗，最後，獲得的音節辨識率約為56.4% (引入語音模型)。除此之外，在我們的系統裡，我們使用KPCA(kernel principal components analysis)方法，去進行基本音節HMM模型分裂，來模擬發音變異現象。

In the thesis，a basic spontaneous Mandarin speech recognition is established. The study focuses on the acoustic modeling for 411 Mandarin base-syllables as well as some special phenomena of spontaneous speech such as particles, uncertain sounds, and paralinguistic phenomena. Performance of the database called MCDC (Mandarin Conversational Dialogue Corpus). Finally, A syllable accuracy rate of 56.4% with adapted language model. In addition，the kernel principal components analysis (KPCA) method is used to split the base-syllable HMM models in order to model the pronunciation variation in our system.