Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/8064
標題: 特定文字的語者辨認之研究
A Study of Text Dependent Speaker Recognition
作者: 張育波
Chang, Yu-Po
關鍵字: Speech recognition;語音辨識;Speaker recognition;dynamic time warping (DTW);Speaker Discriminative Weighting Method;語者辨認;動態時間校準;語者特徵權重法則
出版社: 電機工程學系
摘要: 
傳統的動態時間校準(DTW)演算法,即使當語音樣本資料過大時會面臨到比對耗時之缺點,但仍廣泛地應用在語音辨識上。在本論文中,我們結合了動態時間校準演算法及語者特徵權重法則(Speaker Discriminative Weighting Method)兩種方法,經由模擬實驗之比較,能有效改善上述之情形。
藉由測試音與所建之語音樣本資料庫作比對,可確定語者身份。我們的方法分兩部分;第一部分,先利用特定文字對語者之聲音特徵做一訓練,隨即建構出語者的編碼書(codebook)。第二部分,則是將語者的測試音與語音樣本資料庫做比對以確認語者身份。實驗證明有較佳的辨識率及良好之效能,除此之外,我們所提出之方法花費在辨識率之時間較DTW演算法為省。

Traditional the dynamic time warping (DTW) algorithm was widely used in speech recognition, even though it spends more process time to recognize the voice if we have a large voice database. In the thesis, we combine the traditional the dynamic time warping algorithm and speaker discriminative weighting method which can be effectively improve the performance.
A speaker can be identified by comparing the sample data in a voice character database. There are two steps in our method. In the first part, we created a "codebook" for the speakers to characterize their vocal characteristics using training speech pattern sequences. In the second part, a speaker's voice sample is used to compare the codebook and then to identity the speaker. Experiments show that the new method provides better identification accuracy and is also more reliable, Beside, the propose method spends less time than DTW in recognition rate.
URI: http://hdl.handle.net/11455/8064
Appears in Collections:電機工程學系所

Show full item record
 

Google ScholarTM

Check


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