Analysis and Implementation Of Text Independent Speaker Verification System Using Spectral Characterization Feature And Statistical Modeling

Abstract

Automatic speaker recognition is the deployment of a machine to recognize the speaker newlinefrom a spoken phrase. This technique is one of the most useful and popular biometric recognition newlinetechniques especially in areas where security is a major concern. Automatic speaker recognition newlinehas been an active research area for more than 40 years, and the technology has gradually newlinematured to a state ready for real applications. In the early years, text-dependent recognition was newlinemore studied but gradually the focus has moved towards text-independent recognition because of newlinetheir much wider application field including forensics, teleconferencing, and user interfaces in newlineaddition to security applications. newlineThese systems can operate in two modes speaker identification and speaker verification newlinemode. Speaker identification is the process of determining which registered speaker provides a newlinegiven utterance. Speaker verification, on the other hand, is the process of accepting or rejecting newlinethe identity claim of a speaker. As in any pattern recognition system automatic speaker newlinerecognition systems also have three major components, viz. Feature extraction, speaker modeling newlineand decision making. In the present work first two the subcomponents feature extraction and newlinespeaker modeling of text-independent speaker recognition are studied, and some improvements newlineare proposed for achieving better accuracy and faster processing. newlineCommonly used acoustic features contain both linguistic and speaker information mixed newlinein highly complex way over the frequency spectrum. The solution is to use either better features newlineor better matching strategy, or a combination of the two. Most modern speaker recognition newlinesystems use either Mel-Frequency Cepstral Coefficients ((MFCCs) or Linear Prediction Based newlineCepstral Coefficients (LPCCs). In the present work, in addition to these features other newlinecomplementary spectral features have been used to carry more speaker specific information newlinewhich further strengthens the capability of speaker model. newlineIn the present work, three different