Project update - Sep 18 2017

We achieved some fantastic milestones last week!

Thermal Camera Success

We now have our FLIR Lepton 3 thermal cameras working well enough to start getting more out in the field. For the geeks out there, it turns out that a major source of unreliability was that the Raspberry Pi's SPI clock is tied to the CPU clock. When CPU "turbo mode" kicks in the SPI clock speed also increases, generating an SPI clock rate that is incompatible with the camera. We also have a hunch that clock rate changes during SPI transfers may be causing unreliability.

For now, we are forcing the Raspberry Pi's CPU to stay at its low speed. This gives us a stable SPI clock and allows us to get a reliable stream of thermal video frames off the camera. We have some avenues to explore in terms of avoiding the need to restrict the Raspberry Pi's CPU speed but this approach will do for now.

Given that we now have a workable solution for the camera, we ordered more of them, as well as more Raspberry Pis. Our immediate goal is to gather as much thermal video footage of our target predators as possible for training our machine learning model.

Predator Free 2050 Launch

Our fearless leader spoke at the Christchurch launch of the Predator Free 2050 programme. It was great to hear about all the different approaches to predator eradication that are currently underway and how Predator Free 2050 will help to better organise and fund those efforts.

There was plenty of positive response to what we're doing at the Cacophony Project, both in terms of birdsong analysis and advanced eradication approaches.

Server Infrastructure

We made some important improvements to how the project's server infrastructure is arranged. The preferred way to access the Cacophony Project API is now via HTTPS at Attempts to connect to using HTTP will be redirected to HTTPS.

To continue support for existing deployed devices, we also still offer the API over HTTP if the server is accessed directly via its IP address (i.e. the "old" way). This will be available for a limited time.

To get to the web interface to access uploaded recordings you now need to visit​.nz​/​.

In addition to the publicly visible improvements, we've also made a number of changes on the server to improve robustness and maintainability.

API Documentation

For those of you interested in uploading recordings or interacting with already uploaded data, there's now documentation for the Cacophony Project API. Simply point your web browser at to see it.


Recent bad weather has provided opportunities to test the "hibernation" feature of the Cacophonometer software. When battery levels are low the software will stop recording to conserve power. Once more power is available recordings will resume. The hibernation feature is working well, with one device sleeping for 11 days until more solar power was available.

New functionality has also been added to save recordings onto a local SD card instead of uploading to the cloud. This is to support phones which are out of 3G coverage range, or phones without a SIM card. The idea is that the phones will be left in the field for a while and then retrieved later so that the recordings can be accessed. Desktop software will be developed to handle pulling the recordings off SD cards and then upload them using the Cacophony Project API. 

Machine Learning Intern

Last but not least, we have hired Matthew Aitchison as an intern for the upcoming university break. Matthew has a maths and computing background, and brings a wealth of machine learning knowledge to the table. We have a feeling he's going to make huge progress with our predator identification model.