Acoustic source separation by means of sparse representations

The separation of mixtures of speech signals into the contributing sources, is a challenging theoretical problem of considerable practical significance, in which, despite intensive research in the last two decades, only little progress has been achieved. In this project, we develop algorithms, which are based on sparse signal representations, to resolve the problem of single channel (only one microphone is present) speaker separation. In addition, efficient noise cleaning methods can be formulated based on the same approach.