Convolution blind source separation method based on frequency point correction

Abstract

The invention relates to a convolution blind source separation method based on frequency point correction, and belongs to the technical field of signal processing. The method comprises the steps of finding out similarity among all frequency points through short time Fourier transform (STFT) of a source signal and feature analysis of a corresponding frequency point separation signal; then finding out a frequency point of erroneous blind source separation by setting a threshold (as shown in the description), and then correcting the frequency point by using a correct separation signal at an adjacent frequency point; and finally, carrying out inverse transformation of the signals after sorting and correction, and restoring the separation signal. The method can first separate and restore a speech signal accurately in a low signal to noise ratio environment, achieves excellent actual playing effects, and therefore has a great significance for the study of convolution blind source separation.

Description

technical field [0001] The invention relates to speech signal processing, in particular to a convolution blind source separation problem based on frequency point correction. Background technique [0002] Blind source separation (BSS, blind source separation) is the process of restoring the source signal from the observation signal whose mixing mode is unknown. So far, the field of blind source separation has derived many branches, such as independent and non-independent, linear and nonlinear, underdetermined and non-underdetermined, instantaneous mixing and convolutional mixing, etc., which belongs to the research of instantaneous mixing blind source separation algorithm The most mature, and blind source separation has many potential applications in communication, biological signal processing and other fields, such as robust identification of noise and crosstalk separation. However, in real life, signals are often mixed in a convolutional manner, especially in the case of m...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to solve the frequency point inversion problem caused by the single frequency point blind separation error in the frequency domain blind source separation sorting algorithm, and propose a sorting algorithm for frequency point correction, which solves the problem of voice signal convolution blindness. separation problem

Images