DeepSqueak

February 27, 2019 1:37 PM

Meet DeepSqueak, an algorithm built to decode ultrasonic rat squeaks. Researchers at the University of Washington recently unveiled a remarkable software tool known as DeepSqueak. The program can automatically identify, process, and sort rat and mouse squeaks. It might seem whimsical, but knowing what rodents are squeaking about could be extremely valuable to animal researchers.
posted by Ilira (16 comments total)

Polite young rats don’t squeak until they’re squoken to.
posted by GenjiandProust at 1:42 PM on February 27, 2019 [15 favorites]

They might squeak if they're stuck in a manhole.
posted by Fizz at 1:58 PM on February 27, 2019 [7 favorites]

Extremely valuable for rodent trap researchers, too!
posted by slater at 2:01 PM on February 27, 2019

The next step is clearly autotune.
posted by srboisvert at 2:06 PM on February 27, 2019 [4 favorites]

I can't imagine why rodent trap folks would be interested. I work with speakers that are capable of playing back ultrasonic vocalizations of the sort most commonly made by mice and rats, and they're finicky and very expensive. Most human-made acoustic processors don't make sounds in those frequencies, because no one else is going to hear them. I know this because I've tried to check my mics by recording singing-mouse songs on my phone speakers, which don't even go up that high.
posted by sciatrix at 2:06 PM on February 27, 2019

I mean do you want the Rats of NIMH because this is how we get the Rats of NIMH!
posted by Fizz at 2:11 PM on February 27, 2019 [5 favorites]

Also: singing mice make these kinds of higher-frequency whistling chirps, too, and one of my next projects (if I ever get this goddamn manuscript out) will be digging into them and associating them with social context. There is, for example, a warbling call that is only ever made by male singing mice when there are female singing mice around, as well as

I'm fascinated by the sighing about how long it takes to locate vocalizations in the spectrograms, because our lab is shifting away from recording in short 20-minute intervals immediately after a stimulus and towards 24- or 48-h recordings of singing mice, and... honestly, it never occurred to me to go looking for the vocalizations visually, and I do maybe 80% of our bench behavioral and acoustic work. That just seems so exhausting! My strategy is to contour-map the vocalizations using Matlab routines processing the spectrograms--we used to use the oscillograms looking for things that were sufficiently loud, but no way you can do that in the ultrasonic so I'm transitioning us--then run a bunch of routines to categorize them according to shape (i.e., are they rising in frequency? do they dip? etc) and analyze that. I've never done machine learning, but I suppose I can see how it would be a useful tack--but undergrads? Oof.

I love rats, and I'm way more interested in rat social communication than I am house mice--their social structures are just so much more complex and interesting--so I'll be low-key keeping an eye on this work. I ought to see if they're making a database of rat vocalizations with context--might be fun to practice working with some other types of ultrasonic signals, do some more work with the ultrasonic instead of focusing on my guys' really stereotyped and extreme vocalizations for a little bit.
posted by sciatrix at 2:20 PM on February 27, 2019 [13 favorites]

You might finally be able to get your cat to pay attention to you, too.
posted by GenjiandProust at 2:23 PM on February 27, 2019 [3 favorites]

Extremely valuable for rodent trap researchers, too!

We learned a week or so ago that we have a few mice in the basement. The missus picked up some plug-in Ultrasonic Rodent Repellers, which work a treat. I like very much that the fine print on the back, after touting how economical, easy-to-use, clean and safe the gizmos are, ends with the caution, "DO NOT keep rodent pets such as gerbils, hamsters, guinea pigs, etc. in an ultrasonically protected area." Well, yes.

And while the mice are gone, I wonder if the dog and cat next door to us are experiencing something akin to a car alarm going off for a week solid.
posted by ricochet biscuit at 2:29 PM on February 27, 2019 [2 favorites]

They'll probably discover that rat conversation is fantastically boring; all to do with skittering speed, tail lengths, smelling abilities, and a fair bit about garbage.
posted by Greg_Ace at 2:43 PM on February 27, 2019 [1 favorite]

They'll probably discover that rat conversation is fantastically boring; all to do with skittering speed, tail lengths, smelling abilities, and a fair bit about garbage.

Are they always campaigning for an office of some sort, then?
posted by maxwelton at 2:59 PM on February 27, 2019 [4 favorites]

Bad news: I had never heard of these ultrasonic rodent repellers, but I can't find much, if any evidence that any of them work in rodents. I can find a number of reviews indicating that they don't work on much of anything in field conditions, and a decent recent test-and-review with two species of rats that both confirms the negative data in field tests of these devices and adds a little of its own. (Dogs and cats apparently don't care, but then the only species I could find with an independent test demonstrating any efficacy from these was one that focused on outdoor cats, so take with a grain of salt.)

I mean, I was startled to hear about them because I thought it was such a silly idea, unless you're maybe broadcasting predator vocalizations--which they will eventually ignore if you don't follow it up with some level of harassment--or something of that nature. But as far as I can tell, these devices are emitting sound around 40 kHz, which is both higher-pitched than most of the pain and unpleasant! vocalization ranges of rodents (which tend to be audible to humans), and lower-pitched than many social vocalizations. More, higher-frequency noises tend to degrade quickly throughout an environment and dissipate particularly quickly if they are expected to pass through substrates on the ground; even if the rats and mice find the sound obnoxious and annoying, if there's still good food around and not many predators, they'll eventually habituate. That's how habituation works, I'm afraid; it's the same reason humans live more-or-less cheerfully by airbases.

Also, any of these devices that claims it can do anything worse to a rodent than annoy it is full of shit. These claims I'm seeing that they can kill a rodent or fuck with its digestion are--it's an absurd claim unless you are working with incredibly high decibel ranges, and in that case you should probably worry about human health too as well as the high cost it would demand to produce a device capable of producing noise that loud.
posted by sciatrix at 3:05 PM on February 27, 2019 [4 favorites]

Day one we had a single ultrasound gizmo, which we plugged in in the main floor kitchen, where we had first noticed the evidence. Plugged it in, set up a couple of mousetraps in the kitchen and a couple more in the basement. Next morning, the kitchen mousetraps remained untouched while the basement ones had claimed one victim per. More mousetraps in set in the basement, same results the following morning. Repeat until about day five, when we acquired some more gizmos, which were then deployed downstairs. Not a single mouse in the downstairs traps since then, while the kitchen ones continue to be untouched.

It is not a huge sample size, but I am beginning to see a correlation here. We'll see how it proceeds from here.

TL; dr: Stand back, buddy -- I am doing science.
posted by ricochet biscuit at 3:29 PM on February 27, 2019 [2 favorites]

Unless they're Milford rats. Milford rats are neither seen, nor heard. You can always tell a Milford rat.
posted by SonInLawOfSam at 5:28 PM on February 27, 2019 [1 favorite]

But you can't tell her much.
posted by Daily Alice at 6:08 PM on February 27, 2019 [2 favorites]

It is not a huge sample size, but I am beginning to see a correlation here. We'll see how it proceeds from here.

Science! results: ultimately four mice were taken by mousetraps, always in the basement areas (when there were no ultrasound generators there). The main floor kitchen had evidence of previous mouse activity and it got the first and (at that time only) ultrasonic repellent. No sign of mice living or dead on the main floor after that, and once we tossed a couple of ultrasound plug-in gizmos downstairs, none there either. With all due respect to sciatrix's point that "any of these devices that claims it can do anything worse to a rodent than annoy it is full of shit," I think that is all they do claim to do*: annoy the rodents sufficiently that they go away. Which it seems they have. After three weeks, dead mice in traps have a 100% correlation with areas lacking ultrasound, while areas with ultrasound have not produced a single mouse.

On Saturday, Mrs. Biscuit was moving a mousetrap which she had apparently decided was not set. It was and snapped shut in her hand; although it did not injure her, she was startled. On Sunday our daughter stepped on a trap in the laundry room (fortunately, while wearing shoes) -- she had previously taken note of it but had forgotten it was there and she was likewise startled. The only casualties from traps in the last two weeks have been women's composure, so I think we are prepared to tentatively call this a success.

*On a speculative note, I am not sure I would consider a device that claimed to be able to, say, explode mouse heads through sonic warfare: that seems unnecessarily cruel as well as very messy to clean up.
posted by ricochet biscuit at 9:48 PM on March 10, 2019