Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/18752
標題: 利用K-最近鄰居法辨識中文母音及翹舌音的探討
Using the Method of K- Nearest Neighbor to Recognize Vowel of Isolated Mandarin Word and Investigation of Retroflex
作者: 吳敏男
Wu, Min-Nan
關鍵字: 母音辨識率;Vowel recognition;梅爾頻率倒頻譜係數;K-最近鄰居法;Mel-frequency cepstrum coefficient;K-nearest neighbor
出版社: 統計學研究所
引用: [1] 王小川 (2004),“語音訊號處理”。台北市:全華。 [2] 王國榮 (2000),“Visual Basic 6.0 實戰講座”。台北市:旗標。 [3] 吳明哲,黃世陽 (1998),“Visual Basic 6.0 中文版學習範本”。台北市:松崗。 [4] 鍾靖爵,李宗寶 (2011),“利用共同向量法以及最佳梅爾頻率倒頻譜之特徵辨識特定語者之中文單音”。碩士論文,國立中興大學應用數學研究所,台中。 [5] 羅璟義,李宗寶 (2009),“利用權重式共同向量法於中字彙之特定語者中文單音辨識”。碩士論文,國立中興大學應用數學研究所,台中。 [6] 籃元隆,李宗寶 (2009),“利用權重式多重KNN法於中字彙之特定語者中文單音辨識”。碩士論文,國立中興大學應用數學研究所,台中。 [7] Cover, T. M. and Hart, P. E. (1967). “Nearest Neighbor Pattern Classification”, IEEE Trans. On Information Theory, vol. IT-13, No. 1, pp. 21-27. [8] Gulmezoglu, M. B., Dzhafarov, V. and Barkana, A. (1999). “A novel approach to isolated word recognition”, IEEE Trans. On Speech and Audio Processing, vol. 7, No. 6, pp. 620-627. [9] Keskin, M. Gulmezoglu, M. B. Parlaktuna, O. and Barkana, A. (1996), “Isolated word recognition by extracting personal differences”, in Proc. 6 th Int.Conf. Signal Processing Applications and Technology, Boston, MA, pp. 1989-1992. [10] Tsang-Long Pao. and Wen-Yuan Liao. and Yu-Te Chen. (2007), “Audio-Visual Speech Recognition with Weighted KNN-based Classification in Mandarin Database”, Intelligent Information Hiding and Multimedia Signal Processing, 2007. IIHMSP 2007. Third International Conference on, vol. 1, pp. 39-42.
摘要: 
本篇論文主要是要探討特定語者的1391個國字單音的辨識率,共分成兩部份來進行討論,第一部分是辨識翹舌音子音,第二部分是辨識1391個國字單音的母音,且特徵值分別使用均分倒頻譜及梅爾頻率倒頻譜參數,接著使用K-最近鄰居法進行辨識,而實驗因子有「音框擺盪」、「子音音框數個數」、「音框取樣點數」、「音框的特徵值維度」以及有無使用轉移梅爾頻率倒頻譜。
本次實驗使用包含本人共12組語音資料,最後在各種不同的的特徵參數組合下,翹舌音子音的最高辨識率為98%,1391個國字單音的母音最高辨識率為90.8%。

This study is mainly to recognize 1391 isolated mandarin words for speaker-dependent. I divide the contents into two parts that will be discussed in the following paragraphs. The first part is the recognition of retroflex consonants. The second part is to recognize the 1391 vowel of the isolated mandarin words. We use Mel-Frequency Cepstrum Coefficient (Mfcc) and Uniform Cepstrum respectively to analyze features. Then we use the method of K-nearest neighbor (KNN) for the recognition. Five experimental factors are considered in the paper. That is, “the swing of frame”, “the number of frame”, “the length of frame”, “the dimension of frame” and "the usage of Delta- Mel-Frequency Cepstrum Coefficient”.
The experiment uses 12 groups’ database including mine. Finally, I find that the best recognition rate of retroflex consonants in database is 98% and the best recognition rate of 1391 vowel of the isolated mandarin words is 90.8% in the different combinations of the parameters.
URI: http://hdl.handle.net/11455/18752
其他識別: U0005-0207201214214300
Appears in Collections:統計學研究所

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.