Second Ecoacoustics Webinar: BirdNET Tips and Tricks

We recently held our second webinar in the Open Ecoacoustics Webinar Series, focusing on practical tips and tricks for using BirdNET, the powerful AI tool for bird sound identification.

This session was packed with insights from experts who are pushing the boundaries of ecoacoustic research.

Watch the recording #

The session was recorded and you can watch it below:

Speakers #

Dr Philip Eichinski (QUT) #

Phil opened the talks with a basic introduction to machine learning and BirdNET’s convolutional neural network design. He explained how “embeddings” are generated, how Birdnet uses them for classification, and how user-controlled parameters, such as sensitivity and location-based species lists filter these results.

Dr Nina Scarpelli (CSIRO & Museums Victoria) #

Nina then shared her experiences in developing custom classifiers for target species, including a mammal that BirdNET was not trained on, emphasising the most effective strategies for building a labelled training set for best results. She also shared some insights on using species lists and technical considerations that might influence the results.

An interesting insight was that when using BirdNET off-the-shelf, pelicans make splashing noises when they are near water. This resulted in BirdNET detecting pelicans near water with any water splashing sound.

Callan Alexander (QUT) #

Callan concluded the presentations with practical advice for surveying species assemblages with BirdNET’s own classifier in combination with custom classifiers, including techniques for finding thresholds per species.

Feedback and Next Webinar #

If you’d like to provide feedback on this webinar or suggest topics for future sessions, please fill out our feedback form.

We’ll be hosting one more webinar this year. You can follow the ARDC on EventBrite to be notified of future webinars: ARDC Eventbrite.