Speech enhancement algorithm based on speech existence probability and auditory masking effect

Abstract

The invention discloses a speech enhancement algorithm based on a speech existence probability and an auditory masking effect, and the algorithm comprises the steps: carrying out the preprocessing of an input time-domain speech signal, obtaining a frequency-domain speech signal, and keeping a phase angle; carrying out voice existence probability calculation on the obtained frequency domain signal, and obtaining an estimated noise power spectrum; performing noise masking threshold calculation on the obtained frequency domain signal to obtain a spectral subtraction coefficient value of each frequency point; and performing spectral subtraction in combination with the estimated noise power spectrum and the spectral subtraction coefficient to obtain a pure voice spectrum, and then performing inverse Fourier transform in combination with the reserved phase angle to obtain a pure time domain voice signal. According to the invention, the hearing masking effect of the human ear is utilized, the masking threshold value of the noise signal entering the human ear is calculated, noise estimation is combined, the noise can be eliminated, meanwhile, the perception quality of the voice can be guaranteed as much as possible, and more suddenly-changed peak values are not prone to occurring in the voice signal.

Description

technical field [0001] The invention relates to speech signal enhancement technology, in particular to a speech enhancement algorithm based on speech existence probability and auditory masking effect. Background technique [0002] With the development of technologies such as speech recognition, the field of speech enhancement in its front-end preprocessing has become more and more important. At present, speech enhancement algorithms mainly include spectral subtraction, wavelet transform, Wiener filter and so on. Spectral subtraction can better suppress the noise when the signal-to-noise ratio of the input signal is high, but when the signal-to-noise ratio is low, there are more noise residues. The spectral subtraction method is simple and low in complexity, but the noise estimation deviation is relatively large, and half-wave rectification is used for the negative value obtained after spectral subtraction, which leads to the appearance of "music noise" and seriously affects...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the problem of speech distortion after spectral subtraction due to the large noise estimation deviation in the existing spectral subtraction method, and propose a speech enhancement algorithm based on speech existence probability and auditory masking effect. Denoise, but make the residual noise intensity below the masking threshold, so that the human ear cannot feel the noise, so that the speech distortion can be reduced while eliminating the noise

Images