Multi-element mixed sound signal separation method and device

Abstract

The invention discloses a multi-element mixed sound signal separation method and device. The multi-element mixed sound signal separation method at least comprises the steps of: obtaining the amplitudespectrums and phase spectrums of a mixed sound signal and a real source signal through short-time Fourier transform; training the amplitude spectrum of the mixed sound signal through a generator to obtain an amplitude spectrum of an estimated separated signal; training the amplitude spectrum of the estimated separated signal and the amplitude spectrum of the real source signal through a discriminator, judging whether the amplitude spectrum of the estimated separation signal is real output or not, if yes, obtaining a separation sound signal through the amplitude spectrum of the estimated separation signal and the phase spectrum of the mixed sound signal, and meanwhile, obtaining a generator optimization function to optimize the generator until the maximum number of iterations is reached. According to the multi-element mixed sound signal separation method and device, the intelligibility of the separated sound signal is improved, the generalization ability of the algorithm is good, and the anti-interference performance is very high.

Description

technical field [0001] The invention relates to the technical field of acoustic signal separation, in particular to a method and device for separating multiple mixed acoustic signals. Background technique [0002] Traditional signal processing methods include spectral subtraction, Wiener filtering, etc., all of which estimate the short-time Fourier coefficients of the target sound source under the assumption of the prior distribution of the target sound source and noise. In order to better describe the noise The prior distribution usually assumes that the noise is stationary or slowly changing. However, in practice, this assumption is difficult to satisfy, that is, the separation performance will be seriously degraded under the condition of low signal-to-noise ratio. [0003] With the advent of the era of artificial intelligence and big data, more and more acoustic signals are collected. How to use a large number of real data of known mixed multi-component acoustic signals a...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a multi-component hybrid acoustic signal separation method and device to solve the problems of difficult convergence of the acoustic signal separation model in the prior art, poor practicability and serious energy loss

Images