臺灣博碩士論文加值系統

本篇論文中，我們建立以離散小波轉換為基礎的特徵參數及使用高斯混合模型來代表每位語者的聲學特性。我們實驗所使用的語料分別為TIMIT麥克風語料及本實驗是自行收集的FSC-Tel 1000電話語料。在TIMIT及FSC-Tel 1000實驗中我們得到的最高辨識率分別為99.127%及95.5%。
在建立以離散小波轉換為基礎的特徵參數方面，主要是尋找適當的小波分解樹，並將經過小波分解樹後的小波係數轉換為合適的語者特徵參數，在我們的實驗中發現，不同的小波分解樹對語者識別系統會有很大的影響。在本篇論文中提出幾種有不錯效果的小波分解樹。不過我們尚未找出最佳的小波分解樹。
最後，我們在Windows 2000/XP實做了一個以小波轉換為特徵參數及結合高斯混合模型的語者識別系統。使用者只須錄音25秒即可完成語者登記，而在語者測試時，使用者只須輸入2秒的語音，系統即可辨識出語者身份。在整個系統中花費了大部分的時間在語者的特徵參數擷取，因此，我們往後必須尋找到可以加速特徵擷取的演算法。

In this thesis, we have established a large population speaker identification system which is based on Gaussian Mixture Models using wavelet based features. The corpus for experiments of this thesis are the well-known TIMIT phonetically balance speech database for microphone speech (TIMIT) and a new collected Formosa Speech Corpus for telephone speech (FSC-Tel 1000). We achieve the best speaker accuracy of 99.127% for TIMIT corpus and 95.5% for FSC-Tel 1000 corpus.
About wavelet based features, our objective is to found a suitable wavelet decomposing tree by which the speech waveform could be transformed to the wavelet coefficients. In our experiments, we found that different wavelet decomposing trees play an important role for the accuracy of a speaker identification system. In the thesis we have found several efficient types of wavelet decomposing trees. However the optimal trees have not yet been found.
We also implemented a speaker identification system on Windows 2000/XP in which a user only speaks a segment of speech of about 25 seconds for registering his voice in the system. In real-time testing, each speaker speaks only 2 seconds and then the system can correctly identify him with a high accuracy. In the system the feature extraction consumes the most time. For this reason we must found a good algorithm to speed up the feature extraction.

致謝................................................ v
目錄................................................ vii
圖目錄.............................................. x
表目錄.............................................. xiii
中文摘要............................................ xiv
英文摘要............................................ xv
第一章 緒論........................................ 1
1.1 研究動機........................................ 1
1.2 語者辨識系統概論................................ 3
1.3 研究背景及目的.................................. 6
1.4 研究方法........................................ 8
1.5 章節大要........................................ 10
第二章 語者辨認的基本技術.......................... 11
2.1 語音背景知識.................................... 11
2.2 特徵參數擷取..................... .............. 13
2.2.1 倒頻譜參數................................. 13
2.2.2 梅爾倒頻譜參數............................. 14
2.2.3 能量對數................................... 16
2.2.4 一階導數及二階導數......................... 16
2.3 高斯混合模型與語者辨識.......................... 17
2.3.1 高斯混合模型簡介........................... 17
2.3.2 模型描述................................... 19
2.3.3 模型參數初始化............................. 21
2.3.4 最佳相似性估測法........................... 24
2.3.5 期望值最大演算法........................... 25
2.3.6 語者識別器................................. 30
第三章 小波轉換為基礎之特徵參數.................... 32
3.1 小波理論簡介.................................... 32
3.2 連續訊號小波轉換................................ 36
3.3 離散訊號小波轉換................................ 39
3.4 小波重建........................................ 42
3.5 小波特徵之設計.................................. 43
第四章 實驗設計與實驗結果.......................... 48
4.1 系統架構與實驗設計.............................. 48
4.2 系統效能的評估.................................. 49
4.3 語音資料庫簡介.................................. 50
4.4 麥克風語者辨識實驗.............................. 52
4.5 電話語者辨識實驗................................ 58
第五章 系統實作.................................... 64
5.1 語者識別系統架構................................ 64
5.2 語者識別系統模組................................ 67
5.3 系統展示........................................ 73
第六章 結論與展望.................................. 78
6.1 結論............................................ 78
6.2 展望............................................ 79
參考文獻............................................ 81

【1】Dugelay, J.-L.; Junqua, J.-C.; Kotropoulos, C.; Kuhn, R.; Perronnin, F.; Pitas, I.,“Recent advances in biometric person authentication,” Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP ''02). IEEE International Conference on , Volume: 4 , 2002.
【2】Campbell, J.P., Jr., “Speaker recognition: a tutorial,”Proceedings of the IEEE , Volume: 85 Issue: 9 , Sep 1997.
【3】Reynolds, D.A., “An overview of automatic speaker recognition technology,” Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP ''02). IEEE International Conference on , Volume: 4 , 2002.
【4】Webb, J.J.; Rissanen, E.L., “Speaker identification experiments using HMMs,” Acoustics, Speech, and Signal Processing, 1993. ICASSP-93., 1993 IEEE International Conference on , Volume: 2 , 27-30 Apr 1993.
【5】Reynolds, D.A., “Large population speaker identification using clean and telephone speech,”IEEE Signal Processing Letters , Volume: 2 Issue: 3 , Mar 1995.
【6】R. Sarikaya, B. Pellom, J.H.L. Hansen, "Wavelet Packet Transform Features with Application to Speaker Identification," NORSIG-98 IEEE Norsic Signal Processing Symposium, pp. 81-84, Vigso, Denmark, June 1998.
【7】Daniel J. Mashao and N. Tinyiko Baloyi, “Improvements in the speaker identification rate using feature-sets on a large population database,”Eurospeech 2001.
【8】鍾偉仁，“語者辨認與驗證之初步研究”，臺灣大學電信工程所碩士論文，民國八十九年
【9】范世明，“高斯混合模型在語者辨識與國語語音辨認之應用”，交通大學電信工程所碩士論文，民國九十年
【10】鄭順德，“不特定語句中量語者辨識系統之設計研究”，中山大學電機所碩士論文，民國九十一年
【11】Liu, C.-H.; Chen, O.T.-C., “A text-independent speaker identification system using PARCOR and AR model,”Circuits and Systems, 2002. MWSCAS-2002. The 2002 45th Midwest Symposium on , Volume: 3 , 2002.
【12】Siew Chan Woo; Chee Peng Lim; Osman, R.,“Development of a speaker recognition system using wavelets and artificial neural networks,”Intelligent Multimedia, Video and Speech Processing, 2001. Proceedings of 2001 International Symposium on , 2001.
【13】Torres, H.M.; Rufiner, H.L.,“Automatic speaker identification by means of Mel cepstrum, wavelets and wavelet packets,” Engineering in Medicine and Biology Society, 2000. Proceedings of the 22nd Annual International Conference of the IEEE , Volume: 2 , 2000.
【14】Hsieh, C.-T.; Lai, E.; Wang, Y.-C., “Robust speech features based on wavelet transform with application to speaker identification,” Vision, Image and Signal Processing, IEE Proceedings- , Volume: 149 Issue: 2 , Apr 2002.
【15】Bovbel, E.I.; Kheidorov, I.E.; Chaikou, Y.A.,“Wavelet-based speaker identification,”Digital Signal Processing, 2002. DSP 2002. 2002 14th International Conference on , Volume: 2 , 2002
【16】Reynolds, D.A.; Rose, R.C.,“Robust text-independent speaker identification using Gaussian mixture speaker models,”Speech and Audio Processing, IEEE Transactions on , Volume: 3 Issue: 1 , Jan 1995.
【17】G. McLachlan, Mixture Models. New York：Marcel Dekker, 1988.
【18】Jeff A. Blimes,“A Gentle Tutorial Of The EM Algorithm And Its Application To Parameter Estimation For Gaussian Mixture And Hidden Markov Models,”Interational Computer Science Institute Aprial 1998.
【19】Jialong_He，http://tiger.la.asu.edu/C_lib.htm
【20】Steven Young,“The HTK Book（Version 3.0）”,Cambridge University,2000.