Comparative analysis of speech enhancement methods

Abstract

In this paper, comparative analysis of various speech enhancement methods in different noisy environments is presented. Speech signal corrupted with car noise and F16 noise is used for analysis. Spectral Subtraction, Iterative Spectral Subtraction, Geometrical Approach and Wiener filtering based methods, TSNR (Two Step Noise Reduction) and HRNR (Harmonic Regeneration Noise Reduction) methods are implemented in frequency domain for speech enhancement. Simulations are done on MATLAB7.3. The HRNR gives a maximum 35.77dB improvement in SNR for car noise and a maximum 30.54dB improvement in SNR for F16 noise at -10dB input SNR level.