Source extraction in audio via background learning

Source extraction in audio is an important problem in the study of blind source separation (BSS) with many practical applications. It is a challenging problem when the foreground sources to be extracted are weak compared to the background sources. Traditional techniques often do not work in this setting. In this paper we propose a novel technique for extracting foreground sources. This is achieved by an interval of silence for the foreground sources. Using this silence interval one can learn the background information, allowing the removal or suppression of background sources. Very effective optimization schemes are proposed for the case of two sources and two mixtures.