Speech Enhancement Algorithm of Binary Mask Estimation Based on a Priori SNR Constraints
Jie Wang, Chengcheng Yang, Linhuang Yan, Manlu Huang, Jinqiu Sang · 2018
A speech enhancement algorithm using binary mask estimation and a priori SNR constraints is proposed. The a priori SNR estimation has a major impact on noise spectrum estimation function, so the MMSE rule is used to modify the a priori SNR through secondary processing to get more accurate estimated noise power spectrum and gain function, which are used to retain noise over-estimated Time-Frequency units and discard noise under-estimated Time-Frequency units. Experiments results show that the proposed algorithm can improve the intelligibility of speech signals for low SNR.