Voice Algorithms Spot Parkinson's Disease

another random user writes "Mathematician Max Little discovered that Parkinson's symptoms can be detected by computer algorithms that analyze voice recordings. Now he is looking for volunteers to contribute to a vast voice bank to help the database to learn even more. He is aiming to record up to 10,000 voices and has set up local numbers in 10 countries around the world."

Re:

[skids]

You know I keep reading about all these cheap noninvasive tests, and then go to the doctor for a checkup and expecting for them to wave some gadget in front of my moutth when I exhale, but... nothing... nothing new at all. Just the usual stuff minus, it seems these days, the ball grabbing.

I guess they are all just stuck in the pipeline.

Other neuromuscular syndromes.

[Ungrounded Lightning]

Now if only we had one of these for AIDS

Ha ha but seriously...

Other neurological and/or muscular syndromes (such as ALS or Demyelinating diseases like MS) might produce detectable, but distinct, signatures.

I am not really channeling Katherine Hepburn?

[peter303]

Warbling in the voice? She had dystardia worsening in old age.

Needs more study

[mcgrew]

Alcoholics undergoing withdrawal show the same shakiness as Parkensons victims, I wonder if an alkie going through DTs would show up as Parkinsons with this test?

Recording

[Anonymous Coward]

"After the tone, please say your full name; your date of birth; your address; and your social security number. And thank you for helping fund Parkinson research. BEEEEP."

Re:

[game kid]

It's easy to miss the part just before the ". BEEEEP." where they say " inc. A subsidiary of Parkinson Marketing Group", because they sped it up 'til it became a blip of white noise.

The insurance companies are going to love this

[Anonymous Coward]

When you call to buy insurance

Obamacare eliminates this in 2014

[peter303]

But R.I.P. June 28, 2012.

Re:

[Sponge Bath]

We will find out this week if that protection will be allowed by SCOTUS.

Re:

[Anonymous Coward]

Don't think you know what the word exaggerate means.

Re:

[kanto]

Will it also detect when Michael J Fox intentionally goes off his meds before going on TV to exaggerate his symptoms?

It's not like the people on Fox News do anything different.

Re:

[kanto]

Bleh, might as well let you in on the joke... supposedly "Michael J. Fox went off his meds before going on TV to exaggerate his symptoms" and then I say "The people on Fox News do the same". I guess I'll have to be less subtle next time so as to not get a partisan insightful mod.

Really exciting

[Greenspark]

This is fascinating! I wonder if this will prove to not only identify patients currently expressing symptoms, but have a predictive quality for determining who will suffer from the disease before other tools can detect it. Having an earlier look at patients and charting the developments in their brain would really improve doctors' understanding and probably advance research into treatments quite a bit! if voice can be found to accurately indicate or even predict a disease like parkinsons... what else can it demonstrate? This might be the beginning of a new form of diagnostic sciences.

Re:

[UnknownSoldier]

You mean like this?

Analyzing Speech to Detect Financial Misreporting
www.hbs.edu/units/am/pdf/HMV010410.pdf

How a person talks can already be used to gauge the amount of respect they have for themselves and one another.

Re:

[Igot1forya]

I've been doing something like this for years - I can tell you within a small degree of error if someone is retarded simply by what they say rather than how they say it. For example, many will draw this conclusion just by reading my comments :)

Re:

[Ungrounded Lightning]

It used to be fun to analyze public speaking in the frequency domain and find the 8 to 12 hertz oscillations. The sound equipment used in the stages at the White House appear to remove these in real time these days, though.

Or you may be observing that pathological liars aren't stressed when they lie.

Not News

[Anonymous Coward]

This is great technology, but it isn't news, no matter what the BBC thinks.

UC Irvine's machine learning repository has had a set of Parkinson's and non-Parkinson's voice prints [uci.edu] up since 2008. (Astute readers will note that it's based on Max Little's work, as is the BBC article. But it's not news.)

How about Hitler?

[T.E.D.]

It has been speculated that he suffered from Parkingsons [stackexchange.com]. There's certianly plenty of voice recordings of him, covering many years of activity. This ought to be a great way to test both the software and the theories.

ps. Sorry for Godwining the story so early

Re:

[kanto]

There's plenty of maniacal shouting to audiences, but afaik only one taping [youtube.com] of him talking in conversational tones (in case that makes a difference here). Guardian has some background on how it was taped [guardian.co.uk]; it was in Finland and the guy who did it was lucky to live after the Germans noticed what had happened.

Re:

[X0563511]

Those recordings lack the fidelity for such analysis.

Re:

[T.E.D.]

They are really that much worse than the fidelity you get from an answering machine off of a telephone line?

Re:

[citizenr]

You dont need fidelity for most machine learning problems. Algorithms pick up details that humans consciousness misses.

FBI Could probably help

[Igot1forya]

They should contact the FBI and NSA - they have everyone's voice on file already. "Sir, you're under arrest for %OFFENSE%! By the time your trial reaches a conviction you will be in final stages of Parkinson's, my condolences"

How about 80,000

[uiucgrad]

http://storycorps.org/

Called my local number

[Anonymous Coward]

and the voice recording simply said, "or vagina"

On-Once Th-There W-Was...

[WilyCoder]

T-T-T-Today Junior!

Re:

[Harvey Manfrenjenson]

Since this is machine learning, what does this accuracy measure mean? That for a batch of unknown voices, he predicted the right outcome (parkinson or not-parkinson) 86% of the time?

Here is my algorithm with a precision over 99%:

Parkinson( voice_x) = false;

Mod parent up, please. The phrase "86% accuracy" could mean 4 totally different things:

86% of Parkinson's patients are positively flagged by the algorithm (86% sensitivity)
86% of normal patients are NOT positively flagged by the algorithm (86% specificity)
A positively-flagged result implies that you have an 86% chance of having Parkinson's (86% positive predictive value)
A negatively-flagged result implies that you have an 86% chance of not having Parkinson's (86% negative predictive value)

I think that's ri