Filtering environmental sounds using basic audio cues in robot audition

In this article we present an approach for separating robot-directed speech from environmental sounds for applications in robot audition under high noise conditions. We introduce a new framework for audio processing that combines feature extraction and a grouping process to form what we call audio proto objects. These proto objects combine an arbitrary number of audio features in a compact representation that allows a filtering of environmental sounds using relatively simple audio cues like signal energy or segment length. We demonstrate that our system can be a first step towards a selective and adaptive auditory attention in real-world robotics scenarios.