“It is better able to find the calls than we are,” she said. “It can sometimes pick out examples of, say, a glossy black cockatoo that are really far away from the sound recorders. The calls are really soft or really faint or masked by other birds or wind or rain.

“There’s still a place for human expertise. It’s still important that I really understand what all of their sounds mean. But I can spend more time doing the interpretation and less time trying to just find the birds in the sound files.”

Tom Denton, an acoustics expert and Google Australia software engineer, said the model worked by separating the various animals and calls in a recording into different channels, which researchers could then isolate and use to search even larger sets of data.

“When we work with researchers, especially when they’re working with endangered species, they often have questions that are really fine-grained,” he said. “And this sort of search approach lets us surface answers to those questions efficiently without necessarily needing to have a machine-learning person in the loop.

“For example, if you can tell the difference between adult and juvenile calls, you could look at, say, a colony of birds and get a sense within a year of how much of the population is growing and recovering.”

The same model could be used to detect anything from koalas to cane toads, Denton said, and the model is being made open source to support any conservation efforts that could make use of it.

Loading

“Even just talking to one of the students from the biology department here at QUT, who spent months manually annotating data so they could get the start of a classifier – being able to move them to this sort of system is going to accelerate the number of questions that we can answer,” he said.

“There are scarce resources in this space. So if we can multiply the number of questions someone can answer by 10, then we’ve done a great thing.”

Get news and reviews on technology, gadgets and gaming in our Technology newsletter every Friday. Sign up here.