English  |  正體中文  |  简体中文  |  Items with full text/Total items : 43312/67235
Visitors : 2021394      Online Users : 2
RC Version 5.0 © Powered By DSPACE, MIT. Enhanced by NTU/NCHU Library IR team.

Please use this identifier to cite or link to this item: http://nchuir.lib.nchu.edu.tw/handle/309270000/153894

標題: 探討順序比對法對中文母音辨識之影響
Investigation of the Speech Recognition for Mandarin Word by order and non-order method
作者: 吳志凌
Ng, Chi-Leng
Contributors: 李宗寶
統計學研究所
關鍵字: 梅爾頻率倒頻譜係數;K 最近鄰居法;K-Mean Algorithm
Mel-frequency cepstrum Coefficients;K nearest neighbor method;K-Mean Iterative Algorithm
日期: 2013
Issue Date: 2013-11-19 12:09:27 (UTC+8)
Publisher: 統計學研究所
摘要: 本篇論文主要是利用順序比對法探討特定語者在1391個國語單音國字母音之辨識,所使用的特徵值為梅爾頻率倒頻譜係數(Mel-frequency cepstrum coefficient, MFCC),辨識方法為K最近鄰居法(k-nearest neighbor, KNN),與 K-Mean Algorithm (KMA)。
首先主要是利用轉移梅爾頻率倒頻譜係數和梅爾頻率倒頻譜係數求出其特徵值,然後對訓練語音以KMA進行分類,然後比較三種不同的音框比對方式的辨識率,三種音框比對方式分別是有順序、部份順序和沒有順序;有順序是指先把音框按順序排列,然後和其對應到之音框進行比對;部分順序是指將訓練母音按順序分成五組,然後待測語音和其對應到之組別進行音框比對;沒有順序是指每個測試音的音框與訓練音1-25的音框作比對,找出最小的誤差。而且為了分別看出五個聲調的在各方法的表現,我們有進行個別聲調以及全部聲調母音的辨識,而順序比對法在各項辨識中表現較好,單聲調母音的辨識率最高可達到100%,在全部聲調母音的辨識率最高的辨識率為94.5%,平均母音辨識率為88.9%。
This paper mainly discussed the vowel recognition use of sequence comparison method specific language in the 1391 isolation Mandarin words for speaker dependent, using the features Mel-frequency cepstrum coefficient (MFCC), identification method for the k-nearest neighbor(KNN), and K-Mean Algorithm (KMA).
The first is the use of transform Mel-frequency cepstrum coefficient(Delta -MFCC) and MFCC calculated its features. Training the speech to KMA for classification, and then compare the three different ways of the recognition rate, three kinds of frame than on the way are sequential, some order and non- order. Sequential frame is first in sequence, and then, and that corresponds to the frame for comparison. Part of the training vowels sequence is divided into five groups. Testing the voice and its corresponding group for the frame to compare. Non-order refers to each test 1-25 frame and training frame for comparison to find the smallest error. In order to see the five tones in the performance of each method, we have carried out all the individual tones and tone vowel identification, and sequence comparison method performed better in the identification, single-tone vowel recognition rate can reach 100%, the entire tone vowel recognition rate highest recognition rate was 94.5%, the average vowel recognition rate of 88.9%
Appears in Collections:[依資料類型分類] 碩博士論文

Files in This Item:

File SizeFormat
index.html0KbHTML133View/Open


 


學術資源

著作權聲明

本網站為收錄中興大學學術著作及學術產出,已積極向著作權人取得全文授權,並盡力防止侵害著作權人之權益。如仍發現本網站之數位內容有侵害著作權人權益情事者,請權利人通知本網站維護人員,將盡速為您處理。

本網站之數位內容為國立中興大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用。

聯絡網站維護人員:wyhuang@nchu.edu.tw,04-22840290 # 412。

DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU/NCHU Library IR team Copyright ©   - Feedback