International Journal of Advances in Computer Science and Its Applications
Author(s) : MAHESH CHANDRA, SONU KUMAR
In this paper, we study the effect on speaker identification (SI) system when speech data is recorded on two different sensors, a HP Pavilion third generation laptop and a Samsung mobile ( S3770K) both with built-in microphone in parallel in a closed room in noise free environment. The database contains 10 Hindi sentences (50-60 seconds speech) and one english sentence (7-8 seconds speech) of each 39 speakers (26 Male and 13 Female) in a reading style manner. Identification process adopts the methods of feature extraction based on Mel-frequency cepstrum coefficients (MFCC), linear predictive coding (LPC) coefficients. Gaussian mixture model (GMM) is used as a classifier. Our study shows that higher degradation in performance in case of mismatch of sensors during training and testing of data and MFCC performs better during matched conditions, LPC performs better than MFCC in mismatched conditions .