Noise reduction using spectral-subtraction algorithm with over-subtraction and spectral-reservation factors adapted by harmonic properties

Abstract

The power-spectral-subtraction (PSS) algorithm can remove interference noise efficiently by the subtraction of noise power from the power of a noise-interfered signal. However, the performance of this algorithm is not satisfactory for speech communication. This study proposes using an over-subtraction factor adapted by harmonic properties to increase the ability of noise removal. If the value of the over-subtraction factor is large enough, the interference noise can be removed efficiently; meanwhile, denoised speech suffers from serious speech distortion. On the contrary, plenty of residual noise exists when the value of the over-subtraction factor is too small, causing denoised speech to sound annoying to the human ear. How to define the value of this factor is critical to the quality of denoised speech. In addition, musical residual noise can be well reduced by using a spectral reservation factor. We employed the vowel harmonic properties to define the value of over-subtraction and reservation factors of the noisy spectra by using the sigmoid function. This function maps the relation between the values of over-subtraction and reservation factors, as well as the input SNRs. Experiments revealed that the proposed method can improve the performance of the PSS method significantly by increasing the reduction of interference noise and better preservation on weak vowels.