International Journal of Electronics Communication and Computer Engineering
In speech and audio applications, short-term signal spectrum is often represented using Mel-Frequency Cepstral Co-efficient (MFCC) computed from a windowed Discrete Fourier Transform (DFT). Windowing reduces spectral leakage but variance of the spectrum estimate remains high. An extension to windowed DFT is called multi-taper method which uses multiple time domain windows (tapers) with frequency domain averaging. Then detailed statistical analysis of MFCC bias and variance is done. For speaker verification the extracted feature is used to build a model using classifier GMM, which implements likelihood ratio test to decide whether to accept or reject the speaker.