Wearable devices can help deliver higher levels of interactivity and immersion in audio applications. We present an outline for a system to be paired with such devices that features a separator, a scene analyser and a spatial modification and rendering block, then focus on the separator. With the goal of greater realism and immersion in future applications, we modify the Target Sound Extraction (TSE) scheme to output, other than a mixture of target sounds, a mixture of non-target sounds belonging to classes present in the foreground in the scene and residual ambience. We extend two TSE architectures to the new task and show that the extensions achieve better separation results both in TSE, compared to the original architectures, and in the new task, compared to sequential and parallel TSE applications.