Monaural speech enhancement combining deep neural network and convex optimization

• 作者: ZHANG Xiaoyan ZHANG Tianqi GE Wanying BAI Yangliu
• 来源期刊: 声学学报（英文版）
• 年，卷（期）: 2021,（3）
• 页码范围: 460-476
• 页数: 17页
• 语种: 英文

摘要

The accuracy of noise estimation directly affects the quality of speech enhancement algorithm.To improve the noise suppression effect of current speech enhancement algorithm when noise is estimated and effectively solve the unconstrained optimization problem,a time-frequency mask algorithm based on DNN combined with convex optimization is proposed for monaural speech enhancement.Firstly,the power spectra of noisy speech is extracted as the input of DNN.Secondly,the inter-channel correlation factor between noise and noisy speech is taken as the training target of DNN.Next,the objective function of convex optimization is constructed by using the correlation factor obtained from DNN model.Finally,a new hybrid conjugate gradient method combined with convex optimization,is used for iterative processing on an initial mask.The final mask is used to obtain the enhanced speech.Compared with conventional methods,the simulation results show that under different background noise with low SNR,the obtained ratio mask makes the enhanced speech achieve better LSD,PESQ,STOI and segSNR indices,and improves the overall quality of speech and can effectively suppress noise.