April 1 is the International Bird Day that is intended to make us study and preserve various species of birds. SPbU’s graduate Oleg Kenunen has developed a neural network to help those who are amature bird watchers know which bird is singing perching on their windowsills and those who are experts in bird studies identify the numbers of and species of birds in various habitats.

bird ptica ornitology

Last year, the University openly published 8 unique recordings of bird songs that now can be found in the SPBU’s record library that is one of the world’s richest collections of the animal sounds. They are available to anyone to download into the mobile phones: songs of a siberian rubythroat, chaffinch, black-throated loon, nightingale, red-breasted flycatcher, grasshopper warbler, great tit, and rufous-tailed robin. These recordings were made by the Director of the SPbU’s Laboratory of Bio-Acoustics Evgenii Krechmar.

Another SPbU graduate Oleg Kenunen has developed a mobile neural network application to decode bird songs. The project is part of his graduation thesis while he has been completing a retraining programme in Software and Administration of Information Systems. The reason why he chose this field is that he is a ornithologist who defended his thesis on variations of male bird songs of The Eurasian woodcock (Scolopax rusticola) at St Petersburg University.

Bird watching is extremely popular worldwide.

“It is popular in Europe and North America: bird watching is a hobby for every fifth American. However, we don’t have any software that would allow us to identify the species of the birds. Those applications that we have can analyse the images of the birds, decode rather roughly the bird songs, or allow the user to compare the songs with the recordings”, — said Oleg Kenunen.

The software Oleg kenunen has been developing is a mobile application that allows us to record the voice signals and a server that is based on a neural network to decode the bird songs. Although the software can identify only 20 species of birds, it is incredibly accurate (up to 90%). You can always update the library and teach the neural network – the more recordings the system analyses, the more accurate it will be.

The prototype of the application is ready. All you need is the Internet. Still, you can test the software only on tits, as most birds have not migrated yet. In summer, Oleg is planning to have a series of field experiments to reveal the weaknesses of the application and improve them. In several months, the beta-version of the application is going to be available to download.

“The application will be useful for both amatures and ornithologists, — said Oleg. — If we further develop the software, you will be able to record the bird songs in different locations that will enable you to identify the locations of various species and their numbers. Theoretically, you can cover all the forest and make recordings during 24 hours. It is extremely effort-consuming for a ornithologist”.

Still the key stages are ahead: to further develop the algorithm, to further teach the neural network, to make the design of the application, etc. Besides, we need investors and partners to make the application available to the bird watchers worldwide.