Phase-dependent shared deep convolutional neural network speech enhancement method

Abstract

The invention discloses a phase-dependent shared deep convolutional neural network speech enhancement method, which comprises the following steps of: performing time-frequency domain analysis on noisyspeech data and clean speech data by using short-time Fourier transform to obtain dual-channel time-frequency spectrum characteristics of the noisy speech data and the clean speech data respectively,taking the dual-channel time-frequency spectrum characteristics as training samples, building a shared deep convolutional neural network, training the shared deep convolutional neural network by using a training sample, for to-be-enhanced noisy speech data, obtaining dual-channel time-frequency spectrum features of the noisy speech data, inputting the dual-channel time-frequency spectrum featuresinto the shared deep convolutional neural model, calculating and outputting the predicted dual-channel time-frequency spectrum features, and processing the enhanced dual-channel time-frequency spectrum features by using short-time inverse Fourier transform and an overlap-add method to obtain an enhanced speech signal. The method can effectively suppress noise interference in speech signals and enhance the quality of the speech signals.

Description

technical field [0001] The invention relates to the field of digital speech signal processing, in particular to a phase-correlated shared deep convolutional neural network speech enhancement method. Background technique [0002] Language is one of the most important ways for human to communicate. Communicating through voice makes human life simple and efficient; with the development of mobile communication technology and Internet technology, voice technology is applied in calls, smart speakers, voice recognition, smart In various fields such as security, due to the damaging effect of environmental noise on voice, the application of voice technology in some products does not perform well. Therefore, voice enhancement has become a way to improve voice quality and intelligibility and solve product performance problems in actual environments. an important step. [0003] Speech enhancement technology has received widespread attention in the 19th century, and many solutions have ...

AI Technical Summary

Problems solved by technology

[0007] However, most of the current methods based on deep learning are to design network models, process them in the time-spectrum domain, use amplitude spectrum and power spectrum as training data, and directly perform one-stop noise reduction, ignoring the verification of noise reduction results and speech Phase information; many recent studies have shown that phase plays a vital role in restoring speech signals, especially in the case of severe environmental noise and low signal-to-noise ratio, the inaccuracy of phase information has a great influence on the noise reduction effect Impact

Images