On the optimality of ideal binary time–frequency masks

The concept of ideal binary time–frequency masks has received attention recently in monaural and binaural sound separation. Although often assumed, the optimality of ideal binary masks in terms of signal-to-noise ratio has not been rigorously addressed. In this paper we give a formal treatment on this issue and clarify the conditions for ideal binary masks to be optimal. We also experimentally compare the performance of ideal binary masks to that of ideal ratio masks on a speech mixture database and a music database. The results show that ideal binary masks are close in performance to ideal ratio masks which are closely related to the Wiener filter, the theoretically optimal linear filter.