Using beamforming in the audio source separation problem
Nikolaos Mitianoudis, Mike E. Davies · 2003
The problem of separating audio sources observed in a real room environment is a very challenging task, also known as the cocktail party problem. Much work has been presented on audio separation, even in cases of high reverb. However, various problems remain unsolved in a real-world scenario. In this paper, the authors review proposed solutions employing independent component analysis (ICA), discussing possible solutions to various problems that arise during the analysis (i.e. the permutation problem). In particular, the use of beamforming techniques in parallel with the ICA framework is discussed. Finally, some of the open problems in audio source separation are considered.