Acoustic Source Position Estimation Based On Multi-Feature Gaussian Processes
Andreas Brendel, Ingo Altmann, Walter Kellermann · 2019
Gaussian Processes, representing a Bayesian frame-work for regression, were already previously shown to allow effective range estimation in highly reverberant and noisy scenarios from a single pair of microphones when using the Coherent-to-Diffuse Power Ratio as a feature. In this work we investigate how Gaussian Process regression can jointly estimate range and Direction of Arrival by using the Coherent-to-Diffuse Power Ratio and an additional Direction of Arrival estimation feature (e.g., MUSIC) to achieve an estimate of the source position, based on a single concentrated array requiring only two sensors as a minimum.