Feature Matching Techniques for Speaker Recognition

Abstract

Speaker recognition is a stream of biometric authorization which deals with the automatic identification of individual person using some inherent characteristics of that individual. The last stage of this system is the classification of feature templates generated during the previous stage i.e. feature extraction. This classification stage, also known as feature matching, provides the final decision about the speaker under observation. Hence, it is most important to use appropriate feature matching technique to get the accurate result. There are numerous feature matching techniques which can be used for the purpose. The present work provides an analysis of the various feature matching techniques used in the final step of a speaker recognition system. These techniques can be categorized in Statistical techniques, Soft-computing techniques and hybrid techniques. Statistical techniques include: “Vector Quantization (VQ), Gaussian Mixture Model (GMM), Hidden Markov Model (HMM) etc.”, while Soft-computing techniques are “Artificial Neural Network (ANN), Support Vector Machine (SVM) and Fuzzy logic etc.” Hybrid techniques make use of both the above said techniques.