Voice separation effect algorithm of GA_FastICA algorithm

Abstract

The invention discloses a speech separation effect algorithm of a GA_FastICA algorithm. The algorithm comprises the following steps of: assuming that a noise signal is d (n) and a pure speech signal is f (n) in a traditional spectral subtraction method, and acquiring a noisy mixed signal h (n); performing short-time Fourier transform on the noisy mixed signal to obtain a frequency spectrum; and calculating the relationship between a pure speech power spectrum and a mixed voice power spectrum by utilizing gain calculated by the GA algorithm, so that the power spectrum of the pure speech signalis obtained. With the speech separation effect algorithm adopted, the defect that the assumption that the cross term of a traditional spectral subtraction is zero at about 0dB is not established is overcome; the pure speech signal can be restored to the original speech signal through short-time inverse Fourier transform; and the FastICA algorithm is used for speech separation to obtain the separated speech signal. According to the speech separation effect algorithm of the invention, under different signal to noise ratio conditions, time domain waveforms of the separated speech signal are drawn; the correlation coefficient between the separated signal and the original signal is analyzed; the separation effects of GA_FastICA and FastICA, IBM and FOBI algorithms are compared; the GA_FastICA algorithm can effectively separate the speech signal, and a good speech separation effect is achieved in a noise environment.

Description

technical field [0001] The present invention relates to the technical field of smart home equipment, in particular to the voice separation effect algorithm of the GA_FastICA algorithm. Background technique [0002] With the development of smart home devices, it has become an inevitable trend to control smart devices through audio signals, and it also puts forward higher requirements for voice processing technology. In order to better apply speech processing technology to real life, relevant researchers have conducted a lot of research on speech separation technology, speech recognition technology and speech noise reduction technology. The problem of speech separation originated from the famous cocktail party problem. How to construct an auditory processing mechanism similar to that of the human brain, and then separate the focused speech part from the mixed speech signal for the purpose of specific speech recognition. [0003] According to the number of observation signals ...

AI Technical Summary

Problems solved by technology

[0004] The research of the above-mentioned scholars has made great progress in speech separation and speech enhancement technology, but the speech enhancement technology mainly discusses the effect of suppressing noise in a low SNR ratio environment, without considering the separation of the target speech signal, let alone the separation The case of multiple useful voice signals

Similarly, the speech separation problem does not take into account the applicability of the algorithm when the observed signal is mixed with noise

Images