Requirements

Requirements for the Make Your Own Call Recogniser 2024 Workshop

Dataset #

You will need a “search” dataset of unlabelled longer recordings totalling between 100 and 1000 hours. This must be uploaded to Ecosounds ahead of the workshop. Instructions for this can be found here Uploading Audio to Ecosounds

We will pre-process this audio for use in the workshop, so please ensure it’s uploaded before November 28 2024.

Labelled examples #

In the workshop you will build a call recogniser using just a few short labelled examples. If your search dataset is likely to have plenty of examples of your target species, then just having one or two labelled examples is enough. If your search dataset is likely to have very few or maybe no occurrences of the target species, then you should ideally bring a few more, say 10-20 labelled examples.

Open-sourcing your recogniser #

We hope that if you are successful in creating a working recogniser you will make your recogniser available under an open-source licence (Apache 2.0 https://opensource.org/licenses/Apache-2.0) so that others may use it, and for the recogniser to be published in our registry: https://openecoacoustics.org/resources/registry/