This is a low power, low latency audio classifier developed to be deployed on the ESP32 microcontroller within the Arduino framework for the purpose of continuous, real time bird detection.

The classifier is trained on a custom dataset of bioacoustic and environmental audio captured in urban parks and gardens in The Netherlands.

It is published in combination with the Bird Detector Esp32 DIY repo https://github.com/MatteoHMarangoni/bird_detector_esp32

Measured performance in deployment differs fom what is reported on Edge Impulse. Measured deployment accuracy is 82% when using custom electronics for the microphone input and 64% when using a common MEMS microphone. Measured latency in deployment is circa 50 ms per inference (excluding 1 second sampling time), and power draw is 150mA @ 3.3 volt or approximately half a watt.

Species represented in the dataset include:

• Common Chaffinch
• Eurasian Blackcap
• Egyptian Goose
• Rose Ringed Parakeet
• Eurasian Coot
• Eurasian Magpie
• Mallard
• Eurasian Wren
• Eurasian Jackdaw
• Eurasian Jay
• Gadwall
• Great Spotted Woodpecker
• Pigeon & Doves (combined)
• Short Toed Treecreeper
• Carrion Crow
• Grey Heron
• Common Kingfisher
• Canada Goose
• Common Chiffchaff
• European robin
• Eurasian Blue Tit
• Eurasian Green Woodpecker
• Gulls (combined)
• Eurasian Blackbird
• Great Tit

The noise category contains a broad variety of sources present in urban parks. Most intelligible human speech has been filtered out for privacy reasons.

This project is published by Matteo Marangoni and has been developed as part of the Chorusing Symbionts project. https://matteomarangoni.com/Chorusing-Symbionts-page

License:
Creative Commons Attribution 4.0 International (CC BY 4.0).
You are free to use, share, and adapt this dataset and trained model for any purpose with appropriate attribution.