– For 100 plus years, rodents have been a priceless proxy for humans in medical experiments. By one tally, mouse and rat studies have earned around 75 Nobel prizes in health and physiology. Scientists routinely map rodent genes, alter their brains, and measure every inch of their behavior for the sake of experiments. But what we can’t do, at least not easily, is just listen to what they’re saying. Mice and rats are smart, they’re social, they’re chatty, but a lot of that chatter is above the frequency our ears can hear. If we could listen in, we could learn a ton. But for lots of researchers, ultrasonic rodent calls have always been just out of reach.
– Really, people have known about it for 20 years or so, but the amount of publications that come out is really low, and that’s just because of the time investment involved and the cost of all the equipment. – Kevin Coffey is a senior fellow at the University of Washington School of Medicine. And he just co-authored an interesting paper. Basically, he and his collaborator, Russell Marx, built software that can quickly and accurately process rodent calls.
If you have a special microphone to capture high-frequency audio, it can sift through an audio clip, find rodent chatter, and categorize it. – You can just place a microphone over the cage and listen in and let them tell you exactly how they’re feeling. – This could be huge for all kinds of research, so we wanted to try it for ourselves. But first, we needed to find some rat calls.
So Kevin mailed us one of their special microphones. We can feed it into the laptop and get a graph of the audio spectrum in the room. Way down here, this is audible sound. Human ears tend to max out around 20 kilohertz, but rats can squeak from 20 all the way up to 115. We can’t hear any of that, but we can record it. Step two: record a rat. We found a rat adoption agency called Rattie Ratz, and we invited a volunteer named Jennifer and a rat named Buddy to the studio.
Buddy was more comfortable on shoulders than on the table, so recording audio was tricky. But we did our best to tease out some squeaks. We gave Buddy treats, we let him hang out with his friend, and we tickled him. Rats love being tickled. Anyway, we spent a few minutes recording Buddy and then looked at the audio. I don’t really know what I’m looking at here. I think these long even lines in the ultrasonics, I think that’s the whine from the fluorescent lights or the electronics. This did look promising for a second, these big spikes, but you know what that is? It’s him nibbling on those little snacks. Clearly we needed help actually finding the rat calls, so we went to Seattle.
We met with Kevin and Russell to see how the professionals capture rat sounds, and to learn more about their software, and to play with more rats. Kevin and Russell work at the Neumaier Lab here at the University of Washington. They study anxiety, depression, and drug addiction with the ultimate goal of new drugs or therapies for people, and their work relies on rats. – Rats are great because they’re quick to learn. They’re extremely genetically similar to humans, surprisingly similar, to humans. They’ve produced a lot of translatable outcome, so when you study something in rats, it’s got a high likelihood that it’ll translate to humans. – Their setup is similar to ours: a special mic pointed at the rats and a laptop running audio software. We’re looking for calls in a couple areas. Squeaks down here, around 20 kilohertz, happen when rats are nervous or annoyed. When they’re happy, you tend to see calls up in this 50 kilohertz range.
– These ones, though, are higher frequency and they’re shorter, wavier, whistlier — if we were to slow them down — and they make that when they’re happy. They really love the play. Yeah, those are calls, those are definitely calls. – These are tickle-induced calls. – Yeah, yeah, rats love rough and tumble play. – Scientific term. – Yeah. It’s just staring right down the barrel of the lens. – Yeah, it likes it, probably sees himself. – So we captured lots of calls, but for rat or mouse researchers, the tricky bit isn’t actually the recording — it’s the analysis. – When I did it as a PhD student, we did it all by hand. You would record, and then you would listen to, you know, six hours takes 60 hours if you have to slow it down to be able to hear it. We would record hundreds and hundreds of hours of audio, and to go through that took a team of undergrads a year. – There has been software to help, but it’s expensive and the team found it finicky, which is what made Russell start work on new software they call DeepSqueak.
It’s got that name because it uses deep learning software, which was meant for a totally different purpose: self-driving cars. – They’re called convolutional neural networks, and they’re designed to look at a scene and say okay, that’s a stop sign, or that’s a coffee cup, or that’s a pedestrian, or whatever. You could take the spectrogram and audio signal and turn it into a picture and apply the exact same neural networks that people spent so much money and time producing. – So here’s what that looks like in action. First, Russell looked at our audio from Buddy the rat, and just at a glance, picked out some long 20 kilohertz calls.
Sadly, that means Buddy was a little nervous. (high-pitched squeaks) – So that’s slowed down 20 times. – That’s like a whale sound. – It does. (high pitched squeal) – Wow. – Next, we looked at some of those same calls in DeepSqueak, which automatically found Buddy’s squeaks much faster than real time. – And it’s finished, it’s very fast. Now I can go over here and load the calls in this file and take a look. – Okay, so you can sort of see the shape that’s completely isolated from everything else. – We also pulled up some audio of rats drinking sugar water. They love sugar like they love tickling. – They like to make this pattern where they have a shorter note and then some trails above that. (squeaks) – ID’ing rodent calls is valuable on its own, but DeepSqueak has a couple other tricks up its sleeve. Scientists tend to group rat calls into around 20 categories based on the kind of squeak.
But beyond saying they’re positive or negative, not much is known about what they mean. DeepSqueak can’t tell us that, but it can organize every call it sees. That’s what this chart is, it’s a plot of squeaks organized by shape and pitch. – So you can kind of see the calls over here in this corner are all kind of long and wavy, and down here they’re shorter. – DeepSqueak can even tell you if one call often follows another. Russell calls that syntax, which tiptoes toward an even bigger idea, that these aren’t just calls, but maybe something resembling a language. – The most direct scientific answer is that the order of their calls is not random.
– That’s what you’re willing to say. – That’s not enough to say that it is language or not. – Right. – Those big questions demand their own studies, but Kevin and Russell still have full-time jobs studying depression and addiction. DeepSqueak will certainly help with that. If a rat avoids tasty sugar water in an experiment, is that because it’s happily distracted, or because it’s anxious? The squeaks can give clues, but for now, the team just wants more researchers to start capturing chatter. Because with a big enough library of squeaks, someone will crack the big questions. – To get the Rosetta Stone, you know, like what does each call mean? You kind of need a lot of people working on different behaviors and seeing what calls are made.
– From here on out, they’ll be doing their part. – There’s no down side anymore. Yeah, we’re trying to add it to everything, any time we do something to the animals, we listen. – So if you saw this video and were like, “Oh my gosh, I need a pet rat,” you should consider rescuing one. There are lots of small animal rescue and adoption agencies all around the country.
If you happen to live in the Bay Area, you should check out Rattie Ratz, because they’re great. But otherwise, just do some Googling. There is a ton out there. .