Blind signal separation method and structure, voice control system, and electrical appliance assembly

Abstract

The invention relates to the field of voice control, and in particular, to a blind signal separation method and structure, a voice control system, and an electrical appliance assembly. The blind signal separation method comprises: performing noise reduction preprocessing on a detected voice signal, wherein the voice signal is a linear superposition signal of a plurality of simultaneous signal source voice messages (110); constructing, by means of non-Gaussian measurement, a target function for the preprocessed voice signal (120); estimating, by means of an iterative algorithm, an expectation-maximization separation matrix W of the target function (130); and deriving an estimated target separation signal U(n) from U(n)=WX(n), wherein X(n) is a vector constructed by the plurality of signal source voice messages (140). Noise interference in a home appliance working environment causes a reduction in speech recognition rates and affects voice control functions. Therefore, assumption of a linear instantaneous system is made for an actual working environment of home appliances, and ICA-based blind source separation technology is proposed for a linear instantaneous mixing system, achieving noise reduction of a voice signal.

Dependencies are inferred from the claim text.