Model-based Head Orientation Estimation for Smart Devices
summary
A model-based system called HOE estimates head orientation using only two microphone arrays by matching a theoretical voice energy radiation pattern to measured energy ratios, compensating for distance with a one-time room calibration, then disambiguating symmetric candidates by detecting high-vs-low voice frequency radiation pattern cues. Implementation runs on a laptop with two arrays, tested with 10 volunteers in two rooms, and yields a median error 23°, showing practical, low-training orientation estimation that scales without large microphone networks or heavy ML data collection.
AI-written summary · about