A Novel Windowing Technique for Efficient Computation of MFCC for Speaker Recognition

TL;DR

This paper introduces a new windowing method for computing MFCCs that enhances speaker recognition accuracy by incorporating spectral slope and phase information, outperforming traditional and multitaper methods.

Contribution

A novel windowing technique based on DTFT differentiation that improves MFCC computation for speaker recognition.

Findings

Significant performance improvement over baseline Hamming window.

Outperforms recent multitaper windowing techniques.

Mathematically incorporates spectral slope and phase in cepstrum.

Abstract

In this paper, we propose a novel family of windowing technique to compute Mel Frequency Cepstral Coefficient (MFCC) for automatic speaker recognition from speech. The proposed method is based on fundamental property of discrete time Fourier transform (DTFT) related to differentiation in frequency domain. Classical windowing scheme such as Hamming window is modified to obtain derivatives of discrete time Fourier transform coefficients. It has been mathematically shown that the slope and phase of power spectrum are inherently incorporated in newly computed cepstrum. Speaker recognition systems based on our proposed family of window functions are shown to attain substantial and consistent performance improvement over baseline single tapered Hamming window as well as recently proposed multitaper windowing technique.