Dregon

Official code repository of the SPCUP19: DREGON CHALLENGE

Usage

1. Download or clone this repository;
2. Download the development data (dev_flight.zip and dev_static.zip) from the SPCUP19 page on the Dregon Dataset webpage;
3. Unzip the content of the zip files into the code's folder (see next section);
4. Run the main file baseline.m in MATLAB.
5. (optional) Run the file plot_structure in MATLAB to visualize the structures of the microphone array and drone.

Note: baseline.m features hard-coded path for UNIX-like system. For Windows system, a modification of this variables in needed. Note: the baseline code has been tested on MATLAB R2017a.

Files organization

After 3. point of the previous section, the folder should be organized as follows:

• SPCUP19/
  • MBSSLocate/
    • ...
    • ... files and folders of MBSS Locate MATLAB toolbox
  • baseline.m
  • plot_structure.m
  • dev_flight/
    • audio/
    • SPCUP19_dev_flight.mat
  • dev_static/
    • audio/
    • SPCUP19_dev_static.mat

Baseline with MBSS Locate

As baseline MBSS Locate is used. It is an implementation of the state-of-the-art steered response power with phase transform (SRP-PHAT) algorithm (Dibiase et al., 2001). This implementation is freely available, together with 7 other angular spectrum-based localization techniques (Blandin et al., 2012), in a Matlab toolbox named Multichannel BSS Locate here. In this challenge, the generalized cross-correlation with phase transform (GCC-PHAT) is used for estimating the the angular spectrum of each pair of microphones.

The version shipped within this repository is the base one. A full version with example and data is available on the toolbox website as well.

This toolbox is easy to use:

1. a few line of code in the main files (from 107 till 140) set the parameters for the signals (e.g. sampling frequency, frame/block size), the source (e.g. static/moving, single/multiple) the microphones array (e.g. the array geometry and moving/static behaviour) and the localization method (e.g. angular spectrum technique and resolution grid);
2. the function MBSS_InputParam2Struct(...) tokes as input the parameters obove mentioned and return a structure which is passed to the main fuction of the toolbox;
3. MBSS_locate_spec(...) performs the localization, that is the estimation of both the azimuth(s) and elevation(s) of the target source(s) from the multichannel signals.

Note: MBSS Locate require the knownledge of the microphone array.

References

• SPCUP19 resources
• MBSS Locate toolbox
• DiBiase, J. H., A High Accuracy, Low-Latency Technique for Talker Localization in Reverberant Environments using Microphone Arrays, (Ph.D.). Brown Univ, 2000,
• C. Blandin, A. Ozerov and E. Vincent, Multi-source TDOA estimation in reverberant audio using angular spectra and clustering, Signal Processing 92, pp. 1950-1960, 2012.

SPCUP19_FINAL