Blind Source Separation based on Semblance Beamforming

The source separation task has been tackled from different approaches, including beamforming. The idea of exploiting geometric information of the sensors may contribute in convolutive mixture separation context, usually assumed in real scenario. This work proposes a beamformer algorithm for source separation based on the semblance coherence function. The performance of the algorithm for artificially mixed signals is evaluated using a objective intelligibility metric throughout Monte Carlo simulations in two scenarios and different SNR levels. Results are compared with classic techniques: GSS and Delay and Sum, where the proposed algorithm achieves the best performance under no influence of additive noise.