Deep Squeak was designed to sort mice calls. Now it's detecting whales, too By converting sounds to images, scientists can use artificial intelligence to quickly find and assess animals' calls, even deep in the ocean.

A computer program designed to sort mice squeaks is also finding whales in the deep

  • Download
  • <iframe src="https://www.npr.org/player/embed/1102097099/1102097100" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript

A MARTINEZ, HOST:

Artificial intelligence is booming from deep learning to deep fakes and now deep squeak. NPR's Joe Palca looks at how an AI program with a weird name is being used to search for whales.

JOE PALCA, BYLINE: Yes, deep squeak doesn't sound like a whale detector, but we'll get to that. In 2018, Elizabeth Ferguson founded a company called Ocean Science Analytics. It's a company that, among other things, helps people building offshore wind farms track the impact of their projects on marine mammals to make sure the projects aren't harming the whales.

ELIZABETH FERGUSON: Any kind of operations that happen in the ocean require there be some monitoring or mitigation.

PALCA: Now, you could just go out in a boat and look for whales and dolphins. But that doesn't always give you an accurate count.

FERGUSON: Because some species are difficult to see at the surface. Or they spend a long time at depth.

PALCA: So she wanted to find a better way. Enter Kevin Coffey.

KEVIN COFFEY: I am a behavioral neuroscientist.

PALCA: In particular, Coffey is a behavioral neuroscientist who studies the calls rats and mice make when they're stressed. Those calls are different from the sounds they make when they're not stressed. A lot of his projects are long term, so that means listening to many hours of audio to analyze the rodent calls. But he and colleagues at the University of Washington had an idea.

COFFEY: You take the audio signal. You turn it into an image. And then you can see the calls by eye.

PALCA: And recently, artificial intelligence programs have gotten very good at image analysis using what's called deep learning. So Coffey created a program that was good at recognizing the visual representations of the mouse calls and classifying them as stressed or non-stressed - thus, deep squeak was born. Elizabeth Ferguson figured that what works for mice in cages could be modified to work with marine mammals in the ocean.

FERGUSON: So I'm going to share my screen here.

PALCA: Ferguson is showing me on a computer the results of using her modified version of deep squeak on about 2 1/2 hours of audio recorded off the Oregon coast.

FERGUSON: Anything that has a green box around it is something that was detected in the network.

PALCA: The calls are coming from whales and dolphins within a couple miles of the shore. She can see them thanks to her reworked version of deep squeak.

FERGUSON: So you can see that there's definitely a wide range of calls and a high degree of variability in those calls. But it's still done a pretty good job of detecting them.

PALCA: Ferguson recently presented details of her modifications to deep squeak at a meeting of the Acoustical Society of America. But really, is deep squeak the name you want to use for a program that detects whale calls?

FERGUSON: No, we're going to change it.

PALCA: To what?

FERGUSON: So we're going to be calling it deep waves. Should we find something better? Do you have any suggestions?

PALCA: I'm working on it.

Joe Palca, NPR News.

(SOUNDBITE OF QRION'S "YOUR LOVE")

Copyright © 2022 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.