Speaker Recognition System

speaker recognition

Speech signal is basically meant to carry the information about the linguistic message. But, it also contains the speaker-specific information. This thesis deals with a text based speaker identification system; i.e., to find the identity of a person using his/her speech from a group of persons already enrolled during the training phase.

Feature Extractor is the first component in a speaker recognition system. Feature Extraction transforms the speech signal into a compact but effective representation that is more stable and discriminative than the original signal. For this task we will use the Mel Frequency Cepstral Coefficients.

Speaker Modeling and Pattern Matching then is addressed to accomplish the task of speaker recognition. This will be done using Hidden Markov Modeling (HMM) and (DTW).

The most convenient platform for our work is the Matlab environment, since it is a good tool for signal processing, and many of the needed tasks were already implemented in Matlab.

The main purpose of this thesis is to define feature extraction and matching as a whole, which of the several spectral features are best suited for speaker recognition, and to find out the speech that yields the best discrimination between speakers.

Speech Processing is a growing field of research in which applications are being developed in many products today. Some of these applications consist of features such as speech recognition (to recognize what is being said), and speaker recognition (to recognize the person giving the speech).