Text dependent speaker recognition using HTK

Abstract

The objective of the project is to design a text dependent speaker recognition system using HTK, the HTK toolkit is used essentially for speech recognition, after understanding the theory of Hidden Markov Models, and the different processes HTK uses during recognition, the principle of Log likelihood probability, and the language Model are linked together to identify the speaker of an intended sentence chosen by the user and trained by the speaker wanted to be recognized. After several experiments, many threshold options were used, till arrived to the kinds of threshold playing a major role in speaker identification, a method for choosing thresholds has been formulated and it was applied for 10 speakers. The method for speaker identification worked successfully for all speakers, and the goal of recognizing a sentence that belongs to a particular speaker has been realized.