Sound-Based Bird Detection in the Baltics

In the Baltic city of Gdańsk, Poland, 16 women from different backgrounds, from ornithology to data analysis, meet every second week to jointly work on a machine learning project - sound-based bird species detection.

The idea of the projected emerged after the creation of a local chapter of the global organization known as Women in Machine Learning & Data Science. Three fresh data science bootcamp graduates created WiMLDS Trójmiasto (or Tricity, named for the three cities, Gdańsk, Sopot, and Gdynia) to promote data science careers among women and to integrate a small, local community of female data and machine learning enthusiasts. Since the creation, there have been organized network meetings, workshops, discussions, and presentations by local members and guests.

The project began in August and is meant to last 5 months, during which we focus on sound-based bird species detection. We invited 16 women who had prior experience with Python to work together in a series of two-week long sprints. This project was designed to be a collaboration on a real-life problem which machine learning can help to solve, and not a training or a workshop.

Even though we have a technical mentor, we work mostly by exchanging knowledge and experience within small teams (named after birds’ species!) and then on a group forum. Different backgrounds are a great advantage here, as we have gathered knowledge of ornithology, deep neural network data visualization, sound processing and more. Even though it was not designed as a workshop, the project itself has been planned and prepared in advance: from researching bird sound detection technologies and searching for available data to the creation of a calendar of 2-week long sprints.

The project has the typical structure of a data science project including data research and analysis, data preparation, creation of models, analysis of results (or model improvement) and the final presentation. Moreover, there is already an interest in creating a mobile app to be able to use the model in real life situations…for example while hiking in the woods.