Voice signal dereverberation processing method and device, computer equipment and storage medium

Abstract

The invention relates to a voice signal dereverberation processing method and device based on machine learning, computer equipment and a storage medium. The method comprises the steps: acquiring an original voice signal, and extracting amplitude spectrum characteristics and phase spectrum characteristics of a current frame in the original voice signal; extracting a sub-band amplitude spectrum of the amplitude spectrum characteristics, inputting the sub-band amplitude spectrum into a first reverberation predictor, and outputting a reverberation intensity index corresponding to the current frame; determining a pure voice sub-band spectrum of the current frame by using a second reverberation predictor according to the sub-band amplitude spectrum and the reverberation intensity index; and carrying out signal conversion on the pure voice sub-band spectrum and the phase spectrum characteristics to obtain a pure voice signal after reverberation removal. By adopting the method, the reverberation elimination accuracy of the voice signal can be effectively improved.

Description

Technical field [0001] This application relates to the field of communication technology, and in particular to a method, device, computer equipment, and storage medium for processing speech signal de-reverberation. Background technique [0002] With the rapid development of computer communication technology, a voice call technology based on VoIP (Voice over Internet Protocol, IP-based voice transmission) has emerged, which communicates via the Internet to achieve communication functions such as voice calls and multimedia conferences. In VoIP-based point-to-point calls or multi-person online conference calls, reverberation caused by the speaker's distance from the microphone or poor indoor acoustic environment will cause unclear voices and affect the quality of voice calls. By de-reverberation of single-channel speech, the quality of speech calls can be improved. [0003] The traditional single-channel speech de-reverberation technology needs to obtain the historical frame informat...

AI Technical Summary

Problems solved by technology

These prediction methods often rely on the statistical stationarity or short-term stationarity assumptions of speech reverberation components, but cannot accurately estimate early reverberation including early reflections

For communication scenarios where the environment is constantly changing, the reverberation estimation method based on historical frame information is usually used for reverberation estimation. This method has a certain error in reverberation degree estimation, resulting in a relatively low accuracy rate for reverberation cancellation in single-channel speech. Low

Images