Extraction of Ambience Sound from Microphone Array Recordings for Spatialisation

It is often desirable to separate main (primary) sound sources and background sound in microphone array recordings where the microphone signals are a combination of these two components, and then render primary and background sounds using different rendering techniques. Considerable research has investigated the extraction of the primary sources from the mixture using source extraction and localisation techniques, to then re-spatialise the extracted signals using the desired spatial audio reproduction format (for example, loudspeaker panning, Ambisonics or binauralisation). Little research, however, has focused on the separation and spatialisation of the ambience (background) component of the mixture, like secondary sources, diffuse field, or reverberation. This component is crucial for an accurate and realistic reconstruction of the overall 3D audio scene. This paper presents a new approach to extracting a spatialised version of the background field from a set of microphone array signals containing the original sound field mixture. The approach utilises an LMS algorithm to perform the separation between the primary sources and background sound. The fundamental theory of this technique is presented and validated with experimental measurements.