On the limitations of binaural reproduction of monaural blind source separation output signals

In this contribution we analyze the binaural reproduction of an acoustic scene from monaural blind source separation (BSS) outputs for headphone applications. In practice, a perfect separation of the individual sources cannot be achieved, there are always residual components in the BSS outputs which constitute a major problem for binaural reproduction of an acoustic scene. We derive a necessary condition for the required signal-to-interference-ratio (SIR) at the BSS outputs so that a binaural reproduction of the source signals is possible up to a given tolerable error based on a simplified free-field model. The theoretical findings are verified by simulations with speech signals.