Adaptive blind audio signal separation on a DSP
Jakob van de Laar, Emanuël A. P. Habets, J.D.P.A. Peters, P.A.M. Lokkart · 2001
Abstract — Blind Source Separation (BSS) deals with the problem of separating independent sources from their ob-served mixtures only while both the mixing process and orig-inal sources are unknown. Examples of BSS algorithms em-ployed in acoustical applications can be found among others in audio teleconferencing systems. This paper describes the main ideas and implementation of an Adaptive Blind Signal Separation algorithm. In order to make the real-time im-plementation feasible, the BSS algorithm is based on a sim-plified mixing model (SMM). The input signals are recon-structed by assuming that they are statistically uncorrelated and imposing this constraint on the signal estimates. The nonstationarity of the input signals is used to restrict the set of solutions. The system is realized on a TI TMS320C6701 DSP and is capable of separating two independent simulta-neously occurring audio signals in an ordinary acoustic en-vironment in real-time and in an adaptive way. Finally, the separation performance of the algorithm is evaluated using benchmarks downloaded from the web and own real-world recordings. Keywords—Blind Signal Separation, teleconferencing. I.