Wideband DOA estimation based on multi-channel speech presence probabilities
Mahdi Parchami, Wei‐Ping Zhu · 2014
This work proposes a novel scheme for the direction-of-arrival (DOA) estimation of speech signals by employing speech presence probabilities in the frequency domain. We first obtain an initial estimate of the signal's DOA by conventional beamforming and then use it to compensate for the relative delay across the microphone array signals. The resulting signals are utilized to establish estimates of the speech presence probabilities which are applied on the frequency focusing methods to implement the wideband MUSIC approach in order to determine the final DOA estimate. Simulation results illustrate considerable performance improvement in terms of DOA estimation error in circumstances where the speech signal and background noise have different spectral content. The proposed method is extensively useful in low SNR scenarios and is robust against the choice of the frequency focusing method.