First morepork automatically identified

Tim Hunt is a lecturer at the Centre for Information Technology at Wintec.  Tim has done a lot of work on the software for the Bird Monitor and recently he has been working on automatically detecting moreporks. They are quite common where he lives, and he thought their distinctive sounds may make them easy to identify.

The first part of the project was to create the training data. A Bird Monitor has been captured thousands of recordings. Tim has created some software to make listening to the recordings more effective by automatically identifying events. He has listened to many recordings and manually identified those containing morepork sounds. This provided his training data.

Tim then built a model that was getting about 90% accuracy at identifying moreporks within this training set. This model was then applied to over 5000 sounds identified from recording made by Grant's bird monitor in Akaroa. The model identified over 200 of these as being moreporks, but only 12-19% actually were moreporks (some were hard to tell).

While there is still a lot of room to improve the accuracy, it proves the concept that we can automatically identify birds from the recordings we are making. The morepork model will only improve in accuracy and we can add the capability to identify other birds. It shows the value of starting to collect recordings as soon as possible, with the knowledge that these models can be run over past recordings as they are developed. So eventually you'll be able to get a graph of how individual bird populations have changed in your area over time, showing the benefits of pest control.

Grant was very excited to hear these recordings. He thinks he might hear moreporks once per year maybe. Click on the video below to hear one of the recordings identified by Tim's model. Tim is going to be presenting a paper on his work identifying moreporks at the CITRENZ conference in Nelson in October.

[Note: You can find this paper here]

Leave a comment

Please note, comments must be approved before they are published