Speaker recongnition using neural network

Abstract

Speaker recognition is a biometric operation to determine the identity of the speaker based on speech sample. Speaker recognition mainly divided into speaker identification and speaker verification. In this project, based on signal processing concept; the preprocessing using a voice activated detector (VAD) algorithm is applied and Mel frequency cepstral coefficient (MFCC) is used for the feature extraction, feed-forward artificial neural network is used as a classifier with one hidden layer and two. The system is tested for 11 speakers. For speaker identification, an average identification rate of 94% is achieved with one hidden layer and dropped to 63.1 % in two hidden layers. For speaker verification, an average verification of 100% is achieved.