Modelling a Voice Activated Speaker Identification System using MFCC-Pitch-Formant Vector

摘要

The paper presents the model of an automatic speaker identification system which will recognize users based on their voice. The system will be relatively independent of spoken words but will rely on the voice quality of a user i.e. use speech independent voice recognition. The basic approach was to create a front end system which will identify speech parameters of particular users and create speech feature vectors which will later be used to train a back-propagation neural network for the recognition phase. Mel-frequency cepstrum coefficients and linear predictive coding coefficients have been used, along with Pitch and Formants, for feature extraction. The main area of focus of the paper is to outline the optimum set of speech features which form the most reliable model for an automatic speaker identification system.