Blind extraction of temporomandibular joint sound sources from their anechoic mixtures in a geometrically constrained medium.