A modified spectral subtraction method for speech enhancement based on masking property of human auditory system

This paper addresses the problem of musical noise introduced by conventional spectral subtraction method for speech enhancement. A modified spectral subtraction algorithm based on the masking properties of human auditory system is proposed. In comparison with Virag´s algorithm, the modification of proposed method is made from four aspects. Firstly, VAD (Voice Activity Detection) is substituted by MCRA (Minima-Controlled Recursive Averaging) algorithm to estimate the background noise; Secondly, the masking threshold is calculated based on enhanced speech by multi-band spectral subtraction method; Thirdly, the adaptive parameters of spectral subtraction method is adjusted; Finally, a modified form of parametric spectral subtraction is employed. The performance of the proposed method is evaluated under ITU-T G.160 standard. The results shows that, comparing with the reference algorithms, the proposed method provides acceptable amount of signal-to-noise ratio (SNR) improvement and noise reduction with a little impact on the level of speech. The objective speech quality is improved evidently at the same time.