Fast speech blind source separation method based on frequency point selection under binaural distance

Abstract

The invention discloses a quick speed blind source separation method based on frequency point selection under binaural distance. Firstly two problems in using a hybrid signal covariance matrix determinant as a frequency point selection standard are analyzed and a settlement solution is provided. A preliminary selecting solution is provided for selecting some frequency points for performing frequency domain ICA and ordering, and then finally selected frequency points and unselected frequency points are determined through frequency point screening. Furthermore an amplitude uncertainty problem is settled on the selected frequency points for finishing separation. The unselected frequency points are separated by means of a separating matrix which is established by means of a relative mixing parameter that is extracted by a separating matrix on the selected frequency points. The frequency points with relatively high separating performance can be selected through the preliminary selecting solution and a frequency point screening solution secondary selecting process. A separating algorithm for the unselected frequency points has advantages of no limitation by microphone distance, capability of restricting the frequency points for extracting the relative mixing parameter in a frequency range without space overlapping, and no influence of an amplitude uncertainty problem to a relative damping parameter.

Description

technical field [0001] The invention relates to a fast voice blind source separation method based on frequency point selection under binaural distance. Background technique [0002] The cocktail party problem refers to how we can find the target object from many people who are speaking at a noisy banquet, that is, how to separate specific sound sources only from the sound obtained by linear mixing of some unknown sound sources or all sound sources. Blind source separation (BSS) is a technique to separate the original sound source signal from the mixed sound, and it is used to solve the cocktail party problem. This technology has received attention since the 1990s, and there have been thousands of documents describing the research work in this area. BSS has been widely used in many fields to solve the problem of separating signals, such as speech enhancement in speech recognition, communication field, analysis of ECG and EEG signals. [0003] BSS is a very challenging task...

AI Technical Summary

Problems solved by technology

[0008] (1) Under the condition of small-spacing microphones, the attenuation and delay experienced by the acoustic signal from the same sound source to the two microphones are approximately the same, so the frequency point selection standard based on this model and the separation algorithm for unselected frequency points cannot be applied directly

[0009] (2) When the distance between the microphones increases, spatial aliasing may occur at high frequency points, and the delay parameters or TDOA extracted from the separation matrix of these frequency points are inaccurate, resulting in the separation algorithm of unselected frequency points not working properly

Images