Whitney Yoerger

@wyoerger
| she/her

I'm passionate about mindful birding practices, recording bird vocalizations, and learning more about animal behavior.

Groups

Acoustics

I’m curious about AI software for analyzing nocturnal bird calls, particularly for detecting owl species. I currently use Nighthawk for help with processing my ARU audio files, but it only detects passerines and few other bird families — no owls. Are there other options out there that I should explore?

Carly Batist

@carlybatist | she/her

Science Outreach Lead-Rainforest Connection (RFCx) & ArbimonEcoacoustics, biodiversity monitoring, primates, lemurs

19 October 2023 3:21pm

Hi Whitney! You might try checking out if Arbimon would be helpful for you? You can check out how to get started in our support docs here: https://support.rfcx.org/

Also, depending on what geographic region you're working in and what focal species you're using, BirdNet may also be useful?

Teresa Wicks

19 October 2023 3:21pm

I have a question about Arbimon. I'm working on a project looking for bird use of wet meadow (and associated matrices of habitat). We have two bird lists we've created for BirdNet, a "Master List" of all species (to get an understanding of community data as per input from Indigenous partners) and a "Focal Species List" as per the land managers in put. I will have volunteers doing manual verification + passive listening to attempt to catch false positives and species BirdNet has missed. I recently learned about Arbimon from the Soundscapes to Landscapes project and I'm curious about the audio detector function. Is it detecting spectrograms/sonograms from a provided classifier or does it function similar to BirdNet where we can tell it which species to look for?

Carly Batist

19 October 2023 3:21pm

Hi Teresa,

Thanks for your interest in Arbimon! The platform has a couple different analysis tools that range from unsupervised (like audio event detection & clustering, or AED-C) to semi-automated (pattern matching, random forest). We've got lots more info about each in our support docs here.

The AED-C is an unsupervised machine learning model, so you aren't providing any labels (though the validation page allows you to assign events or clusters as particular species after the fact). The pattern matching is a cross-correlation template matching function where you provide 1 template (example of the species-specific call you're looking for) and the algorithm looks for matches similar to that template. Random forest is a decision-tree-based machine learning model where you provide training clips (presence & absence clips for a species) which the model uses to learn how to classify that species' call. We have developed a number of CNNs (like BirdNet is) but they have more of a regional focus (e.g., one for Kenya, one for western Sumatra, etc. etc.). Right now we run these on the backend, but we are currently working on a public-facing CNN page that we hope to phase in this year.

Hope that helps, but feel free to reach out if you have more questions! You're welcome to also email me directly at [email protected] .

All the best,

Carly

Christos Astaras

@astaras

Wildlife researcher

20 October 2023 2:33pm

Hello Whitney,

Arbimon, as Carly mentioned can be useful, as can the Cluster Analysis tool of Kaleidoscope Pro software. I would also recommend having a look at the new Learning Detector Tool of Raven Pro (link describing it is included below), which taps into the BirdNet database. Assuming you work in parts of the world where most (vocal) bird species are included (e.g. North America, Europe), then you can select the Learning Detector tool to scan for calls of the selected bird species with a threshold that you decide.

All the best,

Christos

Machine Learning Detector – Quick Start Guide

Raven’s new Learning Detector uses TensorFlow machine learning technology to automatically detect and classify sounds. At present the Learning Detector includes BirdNET v2.2, which can detect and classify the sounds of more than 3,000 bird species, as well as machine learning models for the blue whale, North Atlantic Right Whale, and Panamanian katydids. You can also train your own TensorFlow-based machine learning models using Koogu and then integrate them into the Machine Learning Detector.

Cornell Lab of Ornithology - Cornell University

Arky

@arky

Technologist and Visual storyteller focusing on social, conservations issues.

3 November 2023 12:40pm

Interesting! I am also embarking on something similar but thinking of using Raven Pro + BirdNet

Stuart Newson

@snewson

British Trust for Ornithology (BTO)

17 November 2023 1:44pm

Colleagues of mine at the BTO have recently extended the functionality of the Acoustic Pipeline recently (which previously focused on the sound identification of European bats, small mammals and bush-crickets) to also include birds - to support batch processing of recordings. See link below.

This includes a specific classifier for nocturnal breeding birds in Europe - including European owls (also classifiers for nocturnal flight calls for migrant birds - NOCMIG etc.

For people that are happy to share individual clips that are of interest to us, the new bird sound identification functionality is free to use.

I wasn't sure whether you were based in Europe, but if you are, this may be worth trying out.

Audible classifiers

The BTO Acoustic Pipeline is developing a suite of tools for the sound identification of audible bird vocalisations, and of other taxonomic groups.

BTO - British Trust for Ornithology

Owl call detection software

Whitney Yoerger

Groups

Carly Batist

Christos Astaras

Machine Learning Detector – Quick Start Guide

Arky

Stuart Newson

British Trust for Ornithology (BTO)

Audible classifiers