Conservation by computer

Wander into the New Zealand bush, and you’re likely to hear the calls of a variety of native birds. But how do you tell which bird is which, particularly when tui and bellbirds imitate each other, and the trilling of the ruru sounds just like a kiwi? And what does that mean for conservation projects and pest tracking?

Professor Stephen Marsland from the School of Mathematics and Statistics and his research team may have the answer. Over the last four years they have created computer software called AviaNZ that can analyse sound recordings from the New Zealand bush and deduce the abundance of various types of birds.

“The software allows us to monitor the health of bird populations without affecting their behaviours,” Professor Marsland says. “The technology could also monitor pests and predators—by changing the types of recorders you use you might be able to monitor for predator species like rats and stoats that communicate in frequencies that are out of the range of human hearing. It could be an excellent way to gauge the predator stress on a particular area.

“The software can be trained to recognise any sound. As long as the predators are talking to each other and the microphone is close enough, it’ll be able to detect them. We’ve already succeeded in training the software to detect two types of native bats, which are pretty inaudible to humans.”

Professor Marsland’s software analyses spectrogram recordings—visual representations of sound waves—of bush sounds. It then learns the sound waves patterns that represent each bird call and can count how many of those bird calls occur.

Professor Marsland says the software is more accurate than the average, untrained person.

“It’s not as good as an expert human who’s concentrating hard at discerning the different calls, but in any 8-hour recording, most of it is nothing, with no calls at all.

“It’s pretty hard for a human to maintain their concentration listening to that, but the software doesn’t care, and it’s very fast. A 15-minute recording takes it about a minute to process.”

Professor Marsland says the team are now working on being able to tell the direction and distance of the calls from the recorders, and even identifying individual birds.

“We can’t yet tell if the 50 calls are from 5 birds making 10 calls each or 50 birds each calling once, but we are moving toward that.”

The ultimate aim is to turn this into estimates of the abundance of different species.

Professor Marsland and his team have used Science for Technological Innovation National Science Challenge funding to develop a different recorder that uses multiple microphones to mathematically auto-correlate the direction the sounds are coming from and their distance from the microphones. The distance and direction aspects of this project are potentially very useful.

“If you can work out where the calls are coming from, you can start to establish how many birds are doing the talking,” Professor Marsland says.

The team have built a prototype recorder that can successfully estimate the direction of calls but have found estimating the distance from the recorder to be more difficult. They will continue to work on this aspect of the recorder to improve their software and help contribute to conservation efforts in New Zealand.