Speech enhancement model training method and device and speech enhancement method and device

Abstract

The invention relates to the technical field of speech processing, and provides a speech enhancement model training method and device and a speech enhancement method and device. The training method of the speech enhancement model comprises the following steps: acquiring a speech training set, wherein the voice training set comprises noisy voice samples and pure voice samples; acquiring an amplitude spectrum corresponding to the noisy voice sample, inputting the amplitude spectrum into the generation network, and acquiring an enhanced voice amplitude spectrum; acquiring an amplitude spectrum corresponding to the pure voice sample and an enhanced voice amplitude spectrum, and inputting the amplitude spectrum and the enhanced voice amplitude spectrum into a discrimination network to acquire a discrimination result; and adjusting network parameters of the generation network and the discrimination network according to the enhanced voice amplitude spectrum, the amplitude spectrum corresponding to the pure voice sample, the discrimination result and the optimization target, and generating a voice enhancement model. By adopting the method, the performance of the speech enhancement model can be improved, and the speech enhancement effect is further improved.

Description

technical field [0001] The present application relates to the technical field of speech processing, in particular to a speech enhancement model training method and device, and a speech enhancement method and device. Background technique [0002] As one of the mediums of human communication and perception, speech plays an important role in the communication between human beings and the interactive application between human beings and machines. However, in practice, most of the speech signals perceived by users usually contain background noise and interference sound sources. For example, at a noisy dance party, the sounds received by users during communication include not only the target speech of the other speaker, but also Including the background noise of the dance scene and the interfering sound sources of other speakers, it is a typical "cocktail dance" problem. The human ear can clearly judge the content of the target voice of the other speaker by virtue of its unique au...

AI Technical Summary

Problems solved by technology

[0005] However, with the methods of related technologies, since most of them are modeling noise signals, it is assumed that the background noise environment is approximately stationary relative to the region where the target speech exists, so as to use the noise spectrum of the non-speech segment to estimate the noise of the speech segment In addition, it is assumed that the noise signal and the target speech signal are not correlated with each other, and the relationship is additive in the frequency domain

However, in practical applications, the background noise signal does not satisfy these two assumptions, so the enhanced speech signal usually has background noise interference

Images