Blind speech separation : adaptive algorithm and implementation using UMA-16 v2 mic array testbed

Abstract

In an environment where multiple recorded individuals are speaking simultaneously, it is difficult to discern each voice. Therefore, extracting each speech signal from this convoluted mixture is crucial and has several applications. The objective of this work is to perform blind source separation in an adaptive manner. First, we studied the Independent Vector Analysis (IVA) algorithm to fully understand its principle. Then, we modified the algorithm to obtain its adaptive version and added adaptive data whitening to it. Finally, we compared the effects of this whitening on the performance of our algorithm and implemented this method using real signals recorded through an array of microphones