Text Dependent Speaker Identification in Noisy Environment

Abstract

In this paper a comparative study of Linear Prediction Coefficient (LPC) and Mel Frequency Cepstral Coefficient (MFCC) features is presented for text dependent speaker identification in clean and noisy environments. Noisy database was prepared by adding speech and F16 noises to clean database at -5 dB, 0 dB and 10 dB SNR levels. The speaker identification performance with MFCC and LPC features for clean database is 96.65% and 93.65% respectively. MFCC features have also shown better identification rate in presence of both noises at all SNR levels as compared to LPC features. Gaussian mixture model (GMM) was used for training and testing purpose.