Demo paper: Audio object extraction for live sports broadcast

Recent interest in object-based audio systems for cinema opens interesting possibilities to extend the reach of an object-based approach to television broadcast of live events. For events where microphones may be placed close to the source of each sound (e.g. live music) this is relatively straightforward, however for sports events audio objects must be derived from multiple distant microphones. This presents significant challenges in extracting discrete audio objects with coordinate locations from a live event in real time. The research presented here demonstrates the extraction of audio objects from a UK Premier League football (soccer) game using only standard pitch-side microphone positions. Ball-kicks and whistle-blows are extracted in real time with small latency and location information is derived that is used to position the object using 3D coordinates. The resultant sound scene description allows rendering across any reproduction system, including with-height systems. The techniques are equally applicable to other sports with revised object identification templates.