This work examines a semi-blind single-channel source separation problem. Ourspecific aim is to separate one source whose local structure is approximatelyknown, from another a priori unspecified background source, given only a singlelinear combination of the two sources. We propose a separation technique basedon local sparse approximations along the lines of recent efforts in sparserepresentations and dictionary learning. A key feature of our procedure is theonline learning of dictionaries (using only the data itself) to sparsely modelthe background source, which facilitates its separation from thepartially-known source. Our approach is applicable to source separationproblems in various application domains; here, we demonstrate the performanceof our proposed approach via simulation on a stylized audio source separationtask.
展开▼
