BioacousTalks: Custom Machine Learning Models with BirdNET Analyzer GUI

This is a recording of the BioacousTalks presentation from September 19th, 2023. We are joined by Stefan Kahl, creator of the BirdNET algorithm and a researcher at both Technische Universität Chemnitz and the Yang Center.

In this presentation, Stefan Kahl discusses advances in BirdNET, feature embeddings, and transfer learning. Then, he does a demo of the BirdNET Analyzer GUI and demos how we can use this tool to create custom models.

A step-by-step guide for acoustic analysis with BirdNET (including two demo datasets) is available here:

0:00 Introduction
1:18 BirdNET+
5:13 Real-time Acoustic Monitoring
9:30 BirdNET Open-Source
12:11 Feature embeddings
19:53 Examples of using feature embeddings
25:03 Training your own models
27:51 Demo: Install and run the BirdNET Analyzer GUI
34:52 Demo: Train model
39:34 Demo: Run model on multiple files
43:05 Adding a noise folder
47:30 Roadmap for BirdNET GUI
50:30 Demo: Re-run improved model on multiple files
52:35 Demo: Evaluate detections
55:15 Provide feedback

[JMUdukes]

Can this be run thru the command line instead of the GUI?

[ammarhamzi]

Hi, is it possible to use the tflite model in an android app?

[NA-FL]

Great concept, but highly inaccurate identifying birds. I ran BirdNET PI in my backyard here in Ft Myers Florida and not one of the 9 unique bird song species identified was within the continent of North America. A grackle was identified as a pale-winged trumpeter???