A Speech Enhancement System Based on Temporal Modeling Generative Adversarial Networks

Abstract

The invention provides a speech enhancement system based on a time modeling generative adversarial network, belonging to the technical field of speech signal processing, comprising: a data acquisition unit for acquiring a noisy speech signal and down-sampling the noisy speech signal; The signal enhancement unit is used for inputting the noisy speech signal into a generative adversarial network based on time modeling, compressing and extracting the global time domain feature of the speech signal, linking the time domain feature and random noise into a feature vector, and correcting the The feature vector is decoded to obtain an enhanced speech signal. The invention solves the problems of insufficient time dependence and global consideration of speech time domain features, reduces the influence of noise in the speech signal, and improves the hearing quality of the enhanced speech.

Description

technical field

[0001] The invention belongs to the technical field of speech signal processing, and in particular relates to a speech enhancement system based on a temporal modeling generative confrontation network. Background technique

[0002] Speech enhancement is a key technology to improve speech quality and intelligibility, that is, a technology that uses audio signal processing technology to remove noise and extract pure speech signals from an observation signal containing noise. Introducing significant speech distortion remains a formidable challenge.

[0003] In recent years, with the rapid development of artificial intelligence technology and computer processing capabilities, deep learning has become a hot technology in many research fields and has achieved many remarkable research results. Because of the limited performance of traditional speech enhancement algorithms such as Wiener filtering and spectral subtraction, deep learning technology has been introduced...

Images