Efficient window for monolingual and crosslingual speaker identification using MFCC

In this paper an experimental evaluation of the various windowing techniques using mel-frequency cepstral coefficient (MFCC) for monolingual and crosslingual speaker identification is demonstrated. The set of windows presented here allows a tradeoff between main lobe bandwidth and side lobe ripple decay. The speaker identification study is conducted using randomly selected 50 speakers from IITG Multi-variability speaker recognition (IITG-MV) database, MFCC feature and Gaussian mixture model (GMM)-universal background model (UBM) classifier. Speaker identification system based on various windowing techniques shown to have considerably improved performance over baseline Hamming window technique.