Acoustic Source Separation: Fundamental Issues

Acoustic source separation aims at estimating the contribution of each acoustic source from the signals measured at an array of microphones. The measurement signals are usually mixtures of acoustic sources and some noise contributions. When mixtures are assumed to be linear instantaneous mixtures, they can be separated by the method known as blind source separation or independent component analysis. However, in real world applications, the measurement signals are delayed and superimposed sources, i.e., convolutive mixtures of sources. In this paper, we present a fundamental theory and an implementation for blind separation of convolved mixtures of acoustic sources.