METHOD AND APPARATUS FOR SEPARATING OBJECT SOURCE CLASSIFIED SIGNAL USING TWO SENSORS

摘要

PURPOSE: A method and an apparatus for separating an object source classified signal using two sensors are provided to cancel noise in real time from two microphone receiving signals and simultaneously separate signals according to certain N object sources. CONSTITUTION: A signal absence probability calculating unit(11) calculates a global signal absence probability with respect to at least one of the first and second sensor receiving signals according to frames, and calculates a local signal absence probability with respect to each frequency band of a corresponding frame. A signal estimating unit(15) estimates a spectrum vector in which a noise signal is canceled according to each frequency band by the global signal absence probability calculated in the signal absence probability calculating unit. An object source discriminating unit(13) generates an object source classified label vector with a frequency band which belongs in each object source using an attenuation and a delay time generated for a signal value of a frequency band in which it is judged that a signal exists by the local signal absence probability calculated in the signal absence probability calculating unit. An object source classified signal separating unit(17) multiplies the object source classified label vector obtained in the object source discriminating unit by the spectrum vector estimated in the signal estimating unit, and outputs signals separated according to object sources.