Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/18721
標題: 利用KNN最近鄰距法辨識中文單音及爆塞音之探討
Using the Weighted KNN Method to Recognize Isolated Mandarin Word and Investigation of Burst Plug Consonant
作者: 姚信延
Yau, Shin-Yan
關鍵字: Mel-Frequency Cepstrum Coefficient;梅爾頻率倒頻譜係數;K-Nearest Neighbor;K-最近鄰居法
出版社: 統計學研究所
引用: [1] 王小川 (2009)。“語音訊號處理”,台北市:全華。 [2] 本實驗室之資料庫。十一位位特定語者,陳佳妤,李雅琇,吳忠達,林敏琪,鍾靖爵,陳宛余,王進文,洪瑞騰,王為成,吳敏男,陳漢良。 [3] 藍元隆(2009),李宗寶,”利用權重式第K位最鄰近方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。 [4] 李蕙珺(2009),李宗寶,”利用權重式第K位最鄰近方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。 [5] 陳佳妤(2011),李宗寶,”探討梅爾頻率倒頻譜係數之特徵擷取對國語母音之影響” 。碩士論文,國立中興大學應用數學研究所。 [6] 林敏琪(2011),李宗寶探討混合式最佳倒頻譜之特徵擷取對母音之影響。碩士論文,國立中興大學應用數學研究所 [7] 黃世陽,吳明哲,何嘉益,張志成,吳志忠,曹祖聖 (2008)。“Visual Basic 6.0 中文版學習範本”,台北市:松岡。 [8] 王國榮(2000)。”Visual Basic 6.0實戰講座”。台北市:旗標。 [9] 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 [10] Harb, H., and Husseiny, A.H. (2000), “Isolated words recognition using neural networks”, The 7th IEEE International Conference on, 1, 17-20, pp. 349-351.
摘要: 
本論文主是要探討1390國語單字之辨識,本篇論文分成兩部分,第一部分為辨識「ㄅ」﹑「ㄆ」﹑「ㄉ」﹑「ㄍ」爆塞音之辨識,第二部分利用權重式的KNN做全體之辨識。兩部分都是要利用到梅爾倒頻譜係數來求取特徵參數,再利用KNN最近鄰距法辨識。在第一部分的實驗中所討論的參數有「音框擺盪」、「特徵值維度」、「頻率擷取起始點」、「音框取法」、「濾波器取法」;本人最高辨識率達到94.9%。第二部分實驗中所討論的參數有「母音取樣點」、「音框擺盪」、「子母音權重」、「特徵值維度」;本人最高之辨識率74.41%。

This paper is mainly to discuss the speech vowel recognition of 1390 isolated mandarin words for dependence. This paper is divided into two parts.The first part to recognition of burst plug consonant.The second part using weighted KNN to recognition of all words .The two parts all using Mel-Frequency Cepstrum Coefficient(Mfcc) to get features and also using the weighted KNN method to recognize isolated mandarin words.The first part we consider five factors such as the swing of frame and frequency to retrieve the starting or end point and filter emulated and the emulated of frame and the dimension of feature value.The best recognition can be up to 94.9%.In the second part of the experiment have factor that vowel sampling points and swing of frame and frequency to retrieve the starting or end point and the weight of consonant vowel,the best recognition can be up to 74.41%.
URI: http://hdl.handle.net/11455/18721
其他識別: U0005-0907201217331900
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.