Automatic bird detection is coming
Up till now we have only been applying artificial algorithms to thermal videos. This has been successful, making the monitoring of these cameras much easier, saving precious time and improving detection rates.
Soon we hope to have similar benefits available for bird monitoring. At the moment our bird monitors make recordings through the day and upload them to the cloud, where you can listen to them and record what you hear. Personally I find this very tedious. Maybe it's because I'm not an ornithologist. However the nice thing about this is when we develop these algorithms they be run over all the old recordings so you'll be able to see trends from when you first started recording.
The Cacophony Project has recently received funding to secure the services of Chris Blackbourn for a couple of months to develop our first automatic bird monitoring algorithms. Chris first started working on this as an open source contributor and we were impressed with his work. Chris has also been doing some similar work for the South Island Kōkako Trust, helping them analyse recordings in their hunt for the South Island Kōkako.
Chris has three things he's going to try to create in this time:
- A Cacophony Index
- A speech detector
- A bird species identifier
The idea behind a Cacophony Index is to measure the overall volume of birdsong over a period of time. Anecdotally people report that the volume of bird song increases when they remove predators. This will be a way of objectively measuring this. As well as seeing how the birdsong changes at one location we will be able to compare different locations and make a little bit of a regional competition.
Chris is pretty efficient and already has a version of this working and released for the Cacophony team to integrate. So hopefully you'll be seeing this index soon.
Any recording device has the potential to capture human speech and we are just not interested in that or in dealing with the privacy implications. So we've asked Chris to create a speech detector so we can automatically delete any recordings that may contain human speech.
The first version of this is also ready to be integrated. Go Chris!
Bird species identifier
Chris has been experimenting with using a neural network (also known as deep learning) to automatically identify bird species. These structures learn from examples so it is really important to have lots if examples of different bird sounds. The more the better. The Cacophony Project has been building up it's own examples but we're also drawing from other sources, including DOC who has a large collection of tagged bird recordings. If you know of any other examples of data we could use please let us know.
Chris is going to start off with a few easy to identify species that can be identified with a high degree of confidence and then expand the model from there. Once we have something integrated you'll be able to get reports on how individual species are faring in your area. All very exciting.
It will keep getting better
As we release this functionality it's important to remember that it is going to keep getting better. As we get more data and improve our algorithms the capabilities will improve. We will always have the option of re-running new algorithms over old recordings. So start recording now - get your Bird Monitor here.