Speech Enhancement Based on the Combination of Deep Learning and Wavelet Algorithm

In this paper, a method is proposed to enhance the Signal to Noise Ratio (SNR) of speech by combining the wavelet algorithm with deep learning techniques. First, wavelet threshold denoising is introduced for speech enhancement. Second, the deep neural network is proposed to enhance SNR with Ideal Binary Mask. To achieve a better performance, the speech signal is analyzed with these methods with characteristic parameters. Third, these methods compose a novel method to refine the process of speech enhancement. Design efficiency and effectiveness are compared analytically and computationally via numerical experiments, which justifies the superiority of this combination method.