Reservoir computing of (African) bird song

South Africa is one of the top destinations in the world for bird watching. From the total of about 10000 of bird species,850 species are to be found here. Kwazulu-­Natal in particular is one of the richest regions supporting a wide array of different birds, while the Western Cape is another very popular destination. Bird song recordings are available on a large number of internet sites. A rich collection is for example freely available on http://www.xeno­‐canto.org/collection/area/africa. Mobile phone applications for bird watchers are currently being developed, see for example http://www.sasolbirds.co.za/mobile­‐app.php for the “eBirds” of Southern Africa. Some of these applications include also bird song recognition. For example https://itunes.apple.com/en/app/bird­‐song­‐id­‐usa­‐automatic/id681091523?mt=8 is a commercially available application recognizing 156 birds of the Eastern USA. The main intention of this project is to develop a birdsong recognition system for African birds, based on a novel method for time series recognition, namely reservoir computing using nonlinear delay units. Reservoir computing seems to be particularly well suited for such a problem. It is a leading technology for speech recognition and the version based on nonlinear delay units is presently being implemented for the analysis of electro­‐cardiac signals.

Project leader(s):
  • Christian Van den Broeck (Theoretical Physics, University of Hasselt)

Leave a Reply