Code and Algorithms


The baseline sound source localization methods used on the DREGON dataset as described in our IROS 2018 paper [1]  are based on the Multichannel-BSS Locate Matlab toolbox [2].

The following Matlab script should allow one to easily reproduce the in-flight localisation results of the IROS paper: IROS_in_flight_tests.m

We used the following noise-only in-flight recording to estimate the Wiener filter covariance matrices: noise_training_room1.wav

Important Notes:

  • Since the publication of the IROS paper, we decided to rotate the reference frame of the microphone array and flip the azimuth direction. This is why the ground truth azimuth trajectories will not look exactly the same as in the paper. Note that these are just shifted/flipped versions of the same trajectories. The ones obtained in the above script are consistent with the microphone array geometry and all the ground truth position data provided on this website.
  • At the time of writing the IROS paper, the Multichannel-BSS Locate toolbox was still at version 1.3, which did not include Wiener filtering. Results reported in the IROS paper corresponds to v1.3 used with a custom implementation of the Wiener filter. The above script has been upgraded to be compatible with version 2.0 of MBSS Locate which does include Wiener filtering. This explains why slight inconsistencies between the paper’s reported results and the ones obtained using the script may be observed. These slight differences do not significantly change the overall localization results reported in the paper, namely, good performances with white noise sources, poorer performances with speech sources.

IEEE SPCup 2019

The following GitHub includes wrappers to easily use the Multichannel-BSS Locate Matlab toolbox (V2.0) on the IEEE SPCup 2019 data:


[1] Martin Strauss, Pol Mordel, Victor Miguet and Antoine Deleforge, “DREGON: Dataset and Methods for UAV-Embedded Sound Source Localization”, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2018 (submitted).

[2] Romain Lebarbenchon, Ewen Camberlein, Diego Di Carlo, Clément Gaultier, Antoine Deleforge, and Nancy Bertin. “Evaluation of an open-source implementation of the SRP-PHAT algorithm within the 2018 LOCATA challenge.” arXiv preprint arXiv:1812.05901 (2018).

Comments are closed.