Robust remote speaker recognition system based on AR-MFCC features and efficient speech activity detection algorithm

A remote text-independent automatic speaker recognition system has been proposed for communication channel in VoIP applications. The proposed system employs a robust speech feature that uses an efficient speech activity detection algorithm and GMM model. Mel-Frequency Cepstral coefficient (MFCC) is a very useful feature for speech processing in clean conditions but it deteriorates in the presence of noise. Feature extraction framework based on the well known MFCC and autoregressive model (AR) features has been proposed. TIMIT database with speech from 630 speakers has been used in Matlab simulation. The first four utterances for each speaker could be defined as the training set while 1 utterance as the test set. The use of AR-MFCC approach has provided significant improvements in identification rate accuracy when compared with MFCC in noisy environment. However, in terms of runtime, AR-MFCC requires more time to execute than MFCC.