SoylentNews is people
SoylentNews
Academics claim their AI software can detect, with 98.5 per cent accuracy, whether or not someone has caught the COVID-19 coronavirus, just from the sound of their coughing.
To build this software, the MIT team used three ResNet50 models, a popular convolutional neural network designed by Microsoft. They're normally used to process images for computer vision, though in this case they're analyzing audio.
The boffins produced a dataset of 5,320 people, who in April and May submitted audio clips of themselves coughing. Participants also had to fill out a questionnaire that asked if they had caught the coronavirus or not, if they had confirmed this with an official test or not, and what symptoms they had. Thus this experiment relies on the honesty of these human data sources, so bear that in mind.
[...] The results appear promising enough that the team said they are working with a Fortune 100 company to flesh out their model into a fully fledged diagnostic tool.
[...] "The effective implementation of this group diagnostic tool could diminish the spread of the pandemic if everyone uses it before going to a classroom, a factory, or a restaurant."
[...] "The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs," Subirana said.
"This means that when you talk, part of your talking is like coughing, and vice versa. It also means that things we easily derive from fluent speech, AI can pick up simply from coughs, including things like the person's gender, mother tongue, or even emotional state. There's in fact sentiment embedded in how you cough.
(Score: 1, Troll) by fakefuck39 on Monday November 02 2020, @02:46PM (16 children)
"detect just from the sound of their coughing." too bad coughing is not required to have covid. heck, even if your case of covid leads you to coughing, you're not coughing for like the first week or two. So half of the time you have covid that will be serious enough to cough, you're not coughing.
So, these academics can detect it at 98% accuracy? Hmm, I remember learning in 8th grade biology that if it's not a double blind study, and you select candidates that are already coughing, you are a lying faggot. Because to prove is something is accurate, you don't preselect a sample pool that validates your hypothesis.
The bigger question is, why are 7th graders called academics, and why are they writing AI code. More importantly, where are their parents - 7th grade is too early to be taking it up the ass.
(Score: 2) by DannyB on Monday November 02 2020, @02:53PM (7 children)
Those silly doctors think they can detect hernias from having you cough. I wonder if there is any evidence to back it up that idea?
I would use a variation your insightful observation: coughing is not required to have a hernia. Someone with a hernia might not exhibit any coughing symptoms.
So the MIT Lab may simply not know what they're doing. Those idiots. One might even be tempted to think Massachusetts was full of liberals.
Young people won't believe you if you say you used to get Netflix by US Postal Mail.
(Score: 1, Interesting) by Anonymous Coward on Monday November 02 2020, @02:55PM (2 children)
I had covid, but no cough. One of my tentacles swelled up to twice it's normal size, but it is back to normal now.
(Score: 2) by fakefuck39 on Monday November 02 2020, @03:01PM (1 child)
well, you'll be happy to know that they would be able to tell you have covid, even if it did not cause coughs. because they can detect 98.5% of covid.
my little brother had covid. he coughed. only when he smoked weed, daily.
now me, someone who is over 40, quit smoking only 3 years ago after 20 years - I'm sure I'd cough, and they'd identify from the cough that I had covid. However to get their 98.5% accuracy number, they would test me, but not test you. so, as I said, faggots.
(Score: 1, Funny) by Anonymous Coward on Monday November 02 2020, @04:05PM
And were your tentacles swollen too?
(Score: 2) by fakefuck39 on Monday November 02 2020, @02:58PM (3 children)
Oh, they know what they're doing alright. They're publishing a great detection rate without a double-blind study, because "working with a Fortune 100 company." They're trying to get paid.
now, those doctors that can detect hernias by having you cough - they can detect a hernia from your cough, with 98.5% accuracy, even though your hernia has not developed yet, but will?
a hernia requires physical damage to already be there. so no shit you can detect if you have one. covid is in your body and you have zero damage and zero symptoms.
but let me ask you this - would they be able to detect a hernia from a cough without you coughing? because 98.5% ability to identify covid means not just in people coughing.
(Score: 1, Insightful) by Anonymous Coward on Monday November 02 2020, @03:08PM (2 children)
Sounds like you just hate The Science.
(Score: 0) by Anonymous Coward on Monday November 02 2020, @04:06PM (1 child)
Better than not Supporting the Troops.
(Score: 0) by Anonymous Coward on Tuesday November 03 2020, @05:44AM
Right. We need them eggheads to come up with better ways of killing.
(Score: 0) by Anonymous Coward on Monday November 02 2020, @07:54PM (4 children)
I take it you've never coughed voluntarily? What did you do when the doctor moved his stethoscope around on your rib cage and asked you to cough a few times?
I have no problem coughing on demand (or for that matter, on a whim). Sometimes it's useful if it feels like something I ate might be going down the wrong tube (but before any kind of reflex kicks in).
(Score: 2) by fakefuck39 on Monday November 02 2020, @09:10PM (3 children)
Oh, I have. And people who cough on demand or due to a cold stethoscope aren't a part of the test group who produced this 98.5% success. People who cough due to an illness are the test group. They did not take people who don't cough and test them. So if you cough because of the flu, it'll say no covid. So for people who cough, they can tell if the cough is due to covid at 98.5% accuracy. Which is very different than their more general claim of "detects covid 98.5% of the time."
(Score: 0) by Anonymous Coward on Monday November 02 2020, @10:33PM (2 children)
> And people who cough on demand or due to a cold stethoscope aren't a part of the test group who produced this 98.5% success.
From the linked article:
Nothing in there about only selecting people that were coughing because they were sick--they just recorded people coughing. I think you read something else into it that isn't there.
(Score: 0) by Anonymous Coward on Monday November 02 2020, @10:41PM
Digging into the paper at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9208795 [ieee.org]
(Score: 2) by fakefuck39 on Monday November 02 2020, @11:32PM
you state they didn't select people who were sick, then put in a quote saying the opposite. hey, do you think if you put out a request for people to submit for a covid recognition test you might get mostly people who are sick? you know what a double-blind study is - right? it would involve picking random people. they did not pick random people. covid tests are hard to get - in fact it's hasn't even been available to people unless they are showing sick symptoms. so they picked from 100% pool of people who got a covid test. This means the sample pool is people who had some sickness they thought might be covid. There were zero people who had covid and no symptoms. Instead of selecting random people, then giving them the covid test.
(Score: 2) by krishnoid on Monday November 02 2020, @10:27PM (2 children)
The academics know *they* can't detect anything. They feed some audio into a neural network, tweak it so gets better detection rates against a data set, feed it more audio with attached self-reporting data, and read out the accuracy figures.
The "academics" still can't detect anything themselves. If you have a problem with misrepresentation, you go ahead and take issue with the lying faggot neural network producing a 98% detection rate against its data set during a non double-blind study. It may even find your ideas intriguing and wish to subscribe to your newsletter, but I wouldn't presume to associate a probability with that.
(Score: 2) by fakefuck39 on Monday November 02 2020, @11:15PM (1 child)
well, you're no academic that's for sure. I don't have an issue with the neural network. I have an issue with feeding it only data from sick people who are already coughing. I have an issue with claiming "98% covid detection" when it's really "98% covid detection in a pool of people who are coughing from some illness."
They're not detecting covid 98% of the time. They are determining if it's covid or a different illness, in people who are definitely showing sick symptoms. It's like walking into a hospital and hopping beds, and running the test. Yet their lying claim is they picked random people on the street. This has nothing to do with a neural network analyzing audio. It has to do with limiting the input of that network to people showing symptoms of being sick.
something you're not understanding here? maybe start your academics by learning reading.
(Score: 2) by krishnoid on Monday November 02 2020, @11:47PM
After picking up reading in the last 15 minutes, I found this in the paper's abstract [ieee.org]:
So the 98.5% number is for sick people. For asymptomatic people, it's better than that.
(Score: 4, Insightful) by looorg on Monday November 02 2020, @03:34PM (7 children)
Well apparently you just can't cough anymore without having COVID. You can almost see the terror in peoples eyes if you cough in public. Somewhat akin to being some kind of plaguebearer.
(Score: 3, Funny) by ikanreed on Monday November 02 2020, @05:04PM (6 children)
It's like you can't even shoot an AK47 on a crowded street anymore, don't these people know I've loaded it with blanks?
(Score: 0) by Anonymous Coward on Monday November 02 2020, @06:04PM (1 child)
You can't even go out in public and yell Allahu Akbar anymore without getting shot or have people believe you are a terrorist.
(Score: 1, Touché) by Anonymous Coward on Monday November 02 2020, @10:46PM
> You can't even go out in public and yell Allahu Akbar anymore...
Unless, of course, it's dawn and you're in a tower beside the mosque.
(Score: 2) by unauthorized on Tuesday November 03 2020, @03:27PM (3 children)
Terribly sorry for inconveniencing you by contracting the flu, I'll leave my lungs home next time I go out.
(Score: 2) by ikanreed on Tuesday November 03 2020, @04:53PM (2 children)
Your problem is you can't tell difference between when people are judging you versus when people are judging the circumstances they find themselves in because of you.
Now, I do understand your feelings; every pandemic in history has brought with it people making moral judgements on the people who get sick like germs fucking care about your moral character, but to go looking for it in people whose only real concern is their own safety is asshole behavior.
(Score: 2) by unauthorized on Wednesday November 04 2020, @12:36AM (1 child)
Well, I'm glad you are here to tell me what I think. Also, woosh.
(Score: 2) by ikanreed on Wednesday November 04 2020, @02:06AM
And I'm sure glad you're here to tell me I "missed" your unfunny joke.
Your position was clear, the thoughts that led to it are clear, and telling you those thoughts were stupid isn't infringing into some unreasonable psychic domain
(Score: 2) by Mojibake Tengu on Monday November 02 2020, @04:32PM (2 children)
It may even work as intended. But, it does not matter if yes or not.
Selecting, restricting and separating people by simply how they cough is definitely an Apartheid in any interpretation of democracy.
I can predict no just some random riots, but a full scale insurrection, if you do that.
The edge of 太玄 cannot be defined, for it is beyond every aspect of design
(Score: 2) by DannyB on Monday November 02 2020, @04:40PM (1 child)
At least we're not yet separating the horrible people by the shape of their ears. At least, not yet.
Young people won't believe you if you say you used to get Netflix by US Postal Mail.
(Score: 0) by Anonymous Coward on Monday November 02 2020, @05:55PM
I always knew there was something off about you lobe-less freaks!
(Score: 2) by krishnoid on Monday November 02 2020, @09:49PM (3 children)
I knew artificial intelligence was moving forward, but this is kind of incredible. This should significantly accelerate a lot of processes.
Amazing -- of all things, a laboratory that can *think*!
(Score: 0) by Anonymous Coward on Monday November 02 2020, @10:26PM
Color my eye a bit jaundiced, but with the amounts of money flying around for ANYTHING that can quickly and reliably detect COVID, they could have a correlation coefficient of 0.2 and they'd be running to a company to get it out while the money is still warm. Don't forget the patent application. They can probably tweak the software later to make it work, right?
(Score: 0) by Anonymous Coward on Tuesday November 03 2020, @05:22AM
I think it will significantly accelerate the number of bullshit academic who point to the magic box their student constructed and talk in a self-important voice about the significant acceleration of progress. And believe it - after all, the bullshit works doesn't it?
(Score: 3, Insightful) by FatPhil on Wednesday November 04 2020, @12:27PM
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by legont on Tuesday November 03 2020, @01:25AM (3 children)
The cough my friends got was really different. It is very dry and one can't make it wet to achieve a relief of actually coughing something out. It is real bad. My friend told me she never had a cough like this and she actually pissed her panties uncontrollably while coughing. She is denying covid though based on negative test and no other symptoms except general weakness.
So, covid or not, there is an unusual cough out there for sure.
"Wealth is the relentless enemy of understanding" - John Kenneth Galbraith.
(Score: 0) by Anonymous Coward on Tuesday November 03 2020, @04:44AM (1 child)
Time to take another test? This time take an antibody test (assuming it's been weeks since she had the dry cough).
(Score: 0) by Anonymous Coward on Tuesday November 03 2020, @05:47AM
No, she should be immediately tested for vitamin C deficiency. It sounds like she has scurvy. Get her a 5 lb bag of oranges!
(Score: 2) by Taibhsear on Tuesday November 03 2020, @11:32AM
Has she gotten the whooping cough vaccine? If so, maybe check the antibody titers to see if she's still immune...
(Score: 1, Insightful) by Anonymous Coward on Tuesday November 03 2020, @09:35AM (4 children)
If only the people commenting and writing these stories knew the difference between specificity, accuracy, sensitivity, and related measures, we could get some intelligence there too. For something like this, the term "accuracy" is a misleading metric at best and not matching the scientific meaning of accuracy and useless at worse. That said, a quick, highly sensitive, and cost effective screener that people can do at home would be a welcome development.
(Score: 0) by Anonymous Coward on Tuesday November 03 2020, @11:41PM
Turns out there was another, similar study done by a different group, using a different methodology (no AI buzzwords, just good old fashioned signal processing). Just found this link:
https://www.technologyreview.com/2020/10/20/1009332/covid-symptoms-may-hide-in-speech/ [technologyreview.com]
If you are healthy now (or think you are), it's time to save some personal speech samples. Then there will be something for comparison once this analysis becomes available. I'd much rather read a paragraph or two into a mic, than take my oral temperature or blood O2 (both poor and non-specific leading indicators).
(Score: 0) by Anonymous Coward on Wednesday November 04 2020, @12:34AM
I finally got around to reading the paper. Here are the numbers for the curious:
(Score: 2) by FatPhil on Wednesday November 04 2020, @12:41PM (1 child)
Case in point:
type I error
alpha rate
false positive probability
(complement of) specificity
(complement of) selectivity
all represent the same thing. And there are 4 other terms in the "false negative" set too.
I've worked not just in research and teaching, but also in the study of teaching - things like this make you useless cun^H^Hommunicators.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 0) by Anonymous Coward on Wednesday November 04 2020, @08:59PM
You know, I'm not sure if the fact that those are not all equivalent measures buttresses or undermines your point.