Research on Speech Enhancement Model Based on Conditional Deep Convolutional Generative Adversarial Networks
CHU Wei;School of Electrical and Automation Engineering,East China Jiaotong University;
Voice interaction technology is increasingly widely used in real life. Due to the existence of interference,voice interaction technology in real environment is far from satisfactory. In order to improve the performance of speech interaction in real environment,a speech enhancement model based on conditional deep convolutional generative adversarial network(C-DCGAN) is proposed,which adds convolution layers and conditional information to GAN. C-DCGAN uses convolution layers to extract speech features,and uses conditional information to generate high-quality speech. TIMIT database,NOISEX-92 database,Aurora2 database and environmental noise database are used to validate the proposed speech enhancement model. Results showthat compared with spectral subtraction and DNN,the C-DCGAN model improves both PESQ and STOI,and demonstrates that the proposed model can achieve good speech enhancement effect.
【CateGory Index】： TN912.35;TP183