Wing tone

Researchers at Birmingham University are applying pattern recognition techniques to identify the presence of bird species based on their individual song.


Researchers at Birmingham University are applying pattern recognition techniques to identify the presence of bird species based on their individual song.

In collaboration with ecological consultancy Biocensus, the team is hoping to develop an automatic system that will allow ecological experts and bird enthusiasts to estimate the population size of any given bird in a particular region.

Led by Dr Peter Jancovic and Prof Martin Russell, the project will initially focus on the frequency content of birdsongs by the Black Redstart population in Birmingham.

Jancovic said: ‘In principle, the technology is based on analysing how the frequency content of signal varies over time. The recorded signal is split into short-time segments (around 20ms) and for each segment frequency analysis is performed. Once we have a lot of recording for a given bird, a statistical model for the bird is then built to characterise the variations of the frequency in the signal over time.’

The process has been taken from existing speech-recognition technologies, as well as the analysis of time-varying signals used in biomedical data such as ECG signals and the modelling of DNA sequences. The team hopes to build on these systems by developing a process that will counter issues such as environmental noise and sensitivity.

Jancovic added: ‘We hope to improve the technology by employing a combination of several approaches, for instance, using multiple microphones for signal recording, modifying the technology to reflect more appropriately the properties of bird signals in the frequency analysis and modelling, and applying noise-robustness techniques.’

Ellie Zolfagharifard