An Improved Multi-band Speech Enhancement Utilizing Masking Properties of Human Hearing System

This paper discusses the problem of single channel speech enhancement in various noise environments and presents an improved multi-band speech enhancement using masking properties of the human hearing system to address the additive noise and remnant noise, simultaneously. The improved multi-band spectral subtraction (I-MBSS) is used for enhancing the speech degraded by real-world noises. The IMBSS uses an adaptive noise estimation approach to estimate the noise from each band without complicated speech silence detection. The noise is estimated and updated by adaptively smoothing the noisy signal power in each band. The noise estimation technique uses a smoothing parameter which is controlled by a linear function of a-posteriori SNR. Subsequently, the human hearing model is applied in the enhanced speech to compute the noise masking threshold and the subtraction parameters are adjusted according to human perception. The method is tested on speech signals with different noise types at different levels and the results are compared to classical multi-band spectral subtraction algorithm. Speech enhancement performance is evaluated using output SNR and the study of the spectrograms as well as informal listening tests with several types of real-world noises. Based on the analyzed speech signals, the proposed enhancement scheme performs better than then the classicalmulti-band spectral subtraction.