粒子群演算法應用於語者確認系統之研究

PSO Algorithm for Speaker Verification

賴易烽(Yi-fong Lai)

指導教授 ： 莊堯棠

國立中央大學/資訊電機學院/電機工程學系/碩士(2012年)

摘要

本論文將粒子群演算法應用於語者確認系統之向量量化及支撐向量機參數決定兩部分，粒子群演算法是一種模擬鳥群或魚群覓食行為的最佳化方法，在搜尋最佳解的過程中，每顆粒子的位置皆為一組解，依據個體經驗及群體經驗，決定每顆粒子的游動方向，並經由疊代更新位置，搜尋最佳解。粒子群演算法具有容易實現、具記憶性，分散搜尋等優點。
在傳統的語者確認系統上，向量量化大多使用LBG演算法，在疊代的過程中，常會收斂於區域最佳解，因此本論文透過粒子群演算法全域最佳解的搜尋能力，並從實驗的結果獲得比LBG演算法更小的均方誤差，且其收斂速度也優於LBG演算法，確定所提方法的有效性。
此外，語者確認系統在做SVM模型訓練時，如何選擇核函數及其參數值，對於訓練的結果影響重大，本論文利用粒子群演算法來獲得最佳的參數值，實驗的結果顯示，本論文所提之方法，比起傳統上利用grid search來尋找最佳參數值的方法，其相等錯誤率及決策成本函數改善了2.26%和0.0275。

關鍵字

語者確認 ； 粒子群最佳化方法

並列摘要

This thesis proposed method uses PSO algorithm to develop the VQ algorithm and determinate the parameter of SVM. Particle swarm optimization (PSO) simulates social behavior such as birds flocking to a promising position to achieve precise objectives in a multi-dimensional space. PSO performs searches using a population (called swarm) of individuals (called particles) that are updated from iteration to iteration.
The vector quantization (VQ) was a powerful technique in the applications of digital speech compression. The traditionally widely used method such as the Linde-Buzo-Gray (LBG) algorithm always generated local optimal codebook. This thesis proposed method uses PSO algorithm to develop the VQ algorithm. Experimental results showed that the PSO algorithm can provide a better codebook with smaller mean square error (MSE) and less computation time than LBG algorithm.
In the support vector machines (SVM), the model for classification is generated from the training process with the training data. Later on, classification is executed based on trained model. The largest problems encountered in setting up the SVM model are how to select the kernel function and its parameter values. This thesis proposed a method uses PSO algorithm to determinate the SVM parameter. Experimental results showed that the proposed system obtains a 2.26% EER and 0.0275 DCF improvement over the system with grid search.

並列關鍵字

Particle swarm optimization ； Speaker verification

參考文獻

[1] L. Rabiner and B. H. Juang, “Fundamentals of Speech Recognition,” Prentice Hall, vol.103, 1993.

[2] J. P. Campbell and JR., “Speaker recognition: a tutorial,” Proceedings of the IEEE, vol.85, pp. 1437 – 1462, 1997

[3] C. H. Wu and J. H. Chen, “Speech activated telephony email reader (SATER) based on speaker verification and text-to-speech conversion,” IEEE Transactions on Consumer Electronics, vol.43, pp. 707-716, 1997.

[4] T. Jacobs and A. Setlur, “A field study of performance improvements in HMM-based speaker verification,” Second IEEE Workshop on Interactive Voice Technology for Telecommunications Applications, pp.121-124, 1994.

[5] D. Burton, “Text-dependent speaker verification using vector quantization source coding,” IEEE Transactions on Acoustics, Speech and Signal Processing, vol.35, pp.133-143, 1987.