BSS for improved interference estimation for Blind speech signal Extraction with two microphones

Blind source extraction (BSE) as desirable for acoustic cocktail party scenarios requires estimates for the target or interfering signals. Conventional single-channel approaches for obtaining the interference estimate rely on noise and interference estimates during absence of the target signal. For multichannel approaches using multiple microphone signals, a separation of simultaneously active target and interference signals becomes possible if the positions of the target and interfering sources are known. We propose a new method which exploits directional BSS (blind source separation with a geometric constraint) to estimate the interfering speech sources and diffuse background noise jointly and blindly. Herewith we can effectively deal with the underdetermined BSS scenario (more point sources than sensors) in reverberant environments and can even allow for additional babble noise in the background.