CONF Oualil_EUSIPCO_2012/IDIAP A TDOA Gaussian Mixture Model for Improving Acoustic Source Tracking Oualil, Youssef Faubel, Friedrich Magimai-Doss, Mathew Klakow, Dietrich Oualil, Youssef Ed. EXTERNAL https://publications.idiap.ch/attachments/papers/2012/Oualil_EUSIPCO_2012.pdf PUBLIC https://publications.idiap.ch/index.php/publications/showcite/Oualil_Idiap-RR-10-2012 Related documents 20th European Signal Processing Conference 2012 Traditionally, time difference of arrival (TDOA) based acoustic source tracking consists of two stages, more precisely, estimation of TDOAs followed by a tracking algorithm. In general, these two stages are performed separately and presume that (1) TDOAs can be estimated reliably; and (2) the errors in detection behave in a well-defined fashion. The presence of noise and reverberation, however, leads to multimodal TDOA distributions and causes larger errors in the estimates, which ultimately lowers the tracking performance. To counteract this effect, we propose an approach that enhances TDOA estimation by (1) accounting for the multimodal aspect through a Gaussian mixture model and (2) integrating knowledge that has been obtained in the tracking stage. In doing so, this approach tightly couples the two stages. Experimental results on the AV16.3 corpus show that the proposed approach significantly improves the tracking performance compared to various other tracking algorithms. REPORT Oualil_Idiap-RR-10-2012/IDIAP A TDOA Gaussian Mixture Model for Improving Acoustic Source Tracking Oualil, Youssef Faubel, Friedrich Magimai-Doss, Mathew Klakow, Dietrich Direction of arrival estimation Kalman filters microphone arrays tracking EXTERNAL https://publications.idiap.ch/attachments/reports/2012/Oualil_Idiap-RR-10-2012.pdf PUBLIC Idiap-RR-10-2012 2012 Idiap March 2012 Traditionally, time difference of arrival (TDOA) based acoustic source tracking consists of two stages, more precisely, estimation of TDOAs followed by a tracking algorithm. In general, these two stages are performed separately and presume that (1) TDOAs can be estimated reliably; and (2) the errors in detection behave in a well-defined fashion. The presence of noise and reverberation, however, leads to multimodal TDOA distributions and causes larger errors in the estimates, which ultimately lowers the tracking performance. To counteract this effect, we propose an approach that enhances TDOA estimation by (1) accounting for the multimodal aspect through a Gaussian mixture model and (2) integrating knowledge that has been obtained in the tracking stage. In doing so, this approach tightly couples the two stages. Experimental results on the AV16.3 corpus show that the proposed approach improves the tracking performance significantly compared to various other tracking algorithms.