Machine learning for birds conservation

The ‘microfaune’ (or ‘micro-fauna’ in English) project was started last September within our research division at Wazo, the NGO I co-founded. Microfaune aims to develop machine-learning tools for bioacoustics research, in order to foster birds and wildlife conservation in cities.

The goal of the project is to improve the assessment of urban biodiversity with deep learning algorithms. A first step involves the detection of birdsong from audio recordings, made at the Cité Universitaire de Paris using devices provided by the Cornell University Laboratory of Ornithology. The contributions of this project are:

A platform for annotating bird songs (presence or absence)
A model allowing the rapid identification chunking and labelling of bird songs
An open-source labelled database

The Cornell Lab of Ornithology provided us with state-of-the-art recording tools. Independent recording devices can be installed at specific locations of interest in the canopy. A typical device is able to collect approx 250 GO of data every 15 days.

The project, led by Hadrien Jean and his team, was selected for the Fall 2019 and Fall 2020 season of DataForGood, the French incubator for common good.

The code has been made freely available on github, and can be deployed on Google Cloud AI platform and AWS.