Unsupervised Learning based Modified C- ICA for Audio Source Separation in Blind Scenario

Separating audio sources from a convolutive\udmixture of signals from various independent sources is a\udvery fascinating area in personal and professional context.\udThe task of source separation becomes trickier when\udthere is no idea about mixing environment and can be\udtermed as blind audio source separation (BASS). Mixing\udscenario becomes more complicated when there is a\uddifference between number of audio sources and number\udof recording microphones, under determined and over\uddetermined mixing. The main challenge in BASS is\udquality of separation and separation speed and the\udconvergence speed gets compromised when separation\udtechniques focused on quality of separation. This work\udproposed divergence algorithm designed for faster\udconvergence speed along with good quality of separation.\udExperiments are performed for critically determined\udaudio recording, where number of audio sources is equal\udto number of microphones and no noise component is\udtaken into consideration. The result advocates that the\udmodified convex divergence algorithm enhance the\udconvergence speed by 20-22% and good quality of\udseparation than conventional convex divergence ICA,\udFast ICA, JADE