AudioSignal Separation UsingIndependent Subspace Analysis andImproved Subspace Grouping JensWellhausen
Rwth AachenUniversity · 2006
Systems that perform thetaskofautomatic search, retrieval orclassification onaudio signals arebased onautomatic dataexploration algorithms. Formostaudio signals, blind source separation isanimportant preprocessingstep before further classification becomes reliable or evenpossible. Blind source separation isawidefield ofcurrent research, andIndependent Subspace Analysis (ISA) seemstobepromising todealwithmuchkinds of audio signals. Theresult ofanISAperformed onaudio signals isanover-separated setofsubspaces. Inthis paper; newgrouping algorithms fortherecombination of theaudio sources outoftheseparated subspaces arepresented. After areview onISA,different distance measurements forgrouping algorithms arediscussed inthe first part. These distance measurements areusedwithin three different grouping algorithms that arepresented in thesecond part.