An Approach for Speech Enhancement Using Deep Convolutional Neural Network

Abstract

: Speech is a primary and universal medium to communicate with each other. The additive or background noise present in the channel humiliates the signal quality. In order to minimize undesirable background noises, speech enhancement techniques have been introduced. Accordingly, this paper proposes a speech enhancement approach using Deep Convolutional Neural Network (DCNN). At first, the noise signal is appended with the hygienic speech signal and the noisy speech signal is generated. Then, the next step is the framing, in which the Fractional Delta-Amplitude Modulation Spectrogram (FD-AMS) features are extracted from the frames. Finally, the extracted features are provided as the input to the DCNN, which generates the optimized estimation of the speech signal. The proposed method is analyzed using NOIZEUS database based on the metrics, Perceptual Evaluation of Speech Quality (PESQ) and Root Mean Square Error (RMSE). Also, the comparative analysis is performed with the existing speech enhancement techniques. From the results, it is shown that the proposed method obtains maximum PESQ and minimum RMSE than the existing techniques, which shows the superiority of the proposed speech enhancement.