Improving Noise Analysis with the Sound of Silence

Roland Piquepaille writes "Researchers at Rockefeller University have built a mathematical method and written an algorithm based on the way our ears process sound that provides a better way to analyze noise than current methods. Not only is their algorithm faster and more accurate than previous ones used in speech recognition or in seismic analysis, it's also based on a very non-intuitive fact: they know what a sound was by knowing when there was no sound. 'In other words, their pictures were being determined not by where there was volume, but where there was silence.' The researchers think that their algorithm can be used in many applications and that it will soon give computers the same acuity as human ears. Read more for additional references and pictures about this algorithm."

Comments

Eh?

[RedOregon]

What? WHAT? I can't hear you!

Ironic

[neonprimetime]

They were able to visualize the areas in which there was no sound at all. The two researchers used white noise -- hissing similar to what you might hear on an un-tuned FM radio -- because it's the most complex sound available, with exactly the same amount of energy at all frequency levels.

Kinda ironic that the most complex sound is no sound at all?

Re:Ironic

[PitaBred]

Except that's not what they're saying. There is volume. It's just that the levels are even across the board. You can tell the difference between silence and white noise, can't you?
And if you noticed on the histographs of the sounds, that the white noise was just even distribution with small points of silence.
You almost got an easy "insightful". I certainly hope that the mods know better.

Re:Ironic

[neonprimetime]

Agreed ... but don't just rip my posting then ... Shouldn't the name of this /. story be changed then? Per your statements ... "Sound of Silence" is misleading?

Re:Ironic

[Anonymous Coward]

Why would you expect that? a) zero or the absence of a signal is a very hard thing to measure and b) in relativistic QM, for example, the vacuum is not empty --virtual particle/anti-particles boil in and out of existing within the limits of the uncertainty principle. Empty is a relative term; silence is relative.

Re:Ironic

[Omkar]

Ob. Simpsons: How ironic. He's blind, after a lifetime of enjoying being able to see.

Re:Ironic

[gobbo]

This research sends me looking for a research paper I have here somewhere, done during WWII on radio communications. Seems they found that interrupting speech on the radio with bursts of static left it much more intelligible than doing the same with gaps of silence.

Silence is a sorely understudied aspect of communication, in general. It's such a fundamental part of everyday communication, yet little research like this is published.

What Magnasco et al. have done is to move our data processing closer to the kind of predictive observation we think we do with our ears. For instance, the ear habituates to sounds that don't fluctuate much, like that computer fan--they aren't as loud and bothersome after a while. But you notice when the fridge stops. Our hearing is all about dynamic change, and relative silences pattern everything.

So, good on the researchers for tapping into this. I like this quote:

the auditory system is the fastest of the five senses. Researchers credit this discrepancy to a series of lightning-fast calculations in the brain that translate minimal input into maximal understanding. And whatever those calculations are, they're far more precise than any sound-analysis program that exists today.

Basically, they're hinting that auditory processing uses a kind of predictive processing and compression to work quickly. Considering that hearing is basically sloshing fluid tickling tiny hairs, with a couple of transductions in there, it is amazingly fast. Another great thing about this research: using spatial representation for a spatial sense.

Re:Ironic

[gobbo]

silence is relative.

Absolutely. :-)

Really, what we mean by silence is "below the limit of sensitivity." It's all about resting the apparatus in question. And, regarding the Planck frontier: absence of evidence is not evidence of absence. Just because you can't measure it, doesn't mean it's not...

Sound of Silence?

[Red Cape]

I hear dead people..

This is the sound of silence...

[Anonymous Coward]

Nothing to see here. Move along.

p.s. My apologies to Paul Simon.

"When asked to comment on their findings...

[bsdluvr]

...the Rockefeller University researchers remained absolutely silent."

Taggers...

[Anonymous Coward]

Tag this as "rolandpiquepaillespam"

Will the other half buy it?

[DanteLysin]

'In other words, their pictures were being determined not by where there was volume, but where there was silence.'

If only I could sell this theory to my wife.

Re:Will the other half buy it?

[pla]

If only I could sell this theory to my wife.

You won't need to - Women have understood this principle for far longer than we mere males...


"Anything wrong, hon?"

"Nothing."

Re:Will the other half buy it?

[sconeu]

And if you don't know what's wrong, I'm certainly not going to tell you!

The Sound of Silence?

[Anonymous Coward]

Did they use the movie or what?

Seems vaguely similar to dark image analysis

[ishmalius]

This seems almost analogous to dark image analysis in astronomy. This started out as merely taking a photo of nothing, to find the aberrations of the collector. This dark image would be subtracted from the target image, to produce an improved version with a lot of the artifacts removed. It has since grown to basically modeling the environment, and judging what an image of nothing would look like according to the model.

Why?

[spun]

He's stopped linking to his damned blog, so he's not getting undeserved hits. I too was kinda pissed when every third story was a piquepaille post, where he reprinted stories from other sources nearly verbatim interspersed with random pointless comments, but he's not doing that anymore, so what's the big deal?

Re:Why?

[Anonymous Coward]

He's stopped spamming his blog because spamming Slashdot helped him get a full-time job. As far as I'm concerned, as long as he has that job, he should be blacklisted for being a spammer, since he's still reaping the rewards from spamming.

Re:Why?

[Anonymous Coward]

spun (1352) writes: He's stopped linking to his damned blog, so he's not getting undeserved hits.

That's odd because the text "Roland Piquepaille" in the summary is a link to said blog. That is, unless we're talking about different definitions for the words "stopped", "link" and "blog".

(Posting AC because this type of discussion is offtopic)

Re:Why?

[spun]

Scared of losing karma? Whore. Sure, the his name links to his blog, but the story doesn't. Anyone with half a brain could tell that is what I meant, so I think you are deliberately misinterpreting my post in order to make some kind of inane point.

Re:Why?

[spun]

Moron. He never was a spammer. God knows how he got so many stories on Slashdot but it's not like he hijacked open relays and filled the editors inbox with junk mail. There's so many more creative insults you could dish out. Imply that he sucked their dicks: he's a whore! Imply that he bribed them, imply that his dad is their boss, or that he sold the editors drugs. AC, you seriously lack imagination if the worst you can think up is "spammer!"

Re:Why?

[Anonymous Coward]

He's not getting pagerank for it, because Slashdot has added the rel="nofollow" attribute to links from submitters (go view source).

Re:Yes, but...

[PitaBred]

I hear that crap when the TV is on, too. Annoys me like nothing else.
But on-topic, I'd say yes. It's definitely not silence...

Can you hear me now?

[Overzeetop]

Sorry, just took a break from a busy day, and didn't see the obvious useless reply post.

By the time True AI is here...

[ZSpade]

They'll already have all of the senses ready to plug in. Just earlier on slashdot this article was posted about providing the robot sense of touch: article.pl?sid=06/06/11/1656248

Not only that, but there have been numerous articles on the development of electronic eyes. By the time they've got all the kinks worked out in AI they'll already be able to let the new robots sense our world in the way we do. The only thing they're really missing are the senses of smell and taste. I can imagine those won't be nearly as hard to duplicate though. Just sensors that detect the chemical make-up of items or particles the air.

Re:By the time True AI is here...

[Evanisincontrol]

But isn't taste a little more complicated than that? When I eat a cookie, I might say it's delicious. But when you eat the cookie, you think it is dry and tasteless. Who is the ultimate authority on what an object "tastes like"?

Granted, it's easy enough for two people to agree that a lemon is sour instead of sweet, but where exactly is the line between the two? Will you and me (and our Robo-Taster 2000) agree every time that an object is sour instead of sweet? Maybe, if we give it a numerical value, but how would we assign numerical values to a taste? Can I say that my lemon has 63 sour units?

A computer may be able to analyze chemical make-up of an object, but the interpretation of those chemicals is the difficult part. If anyone has any insight into the subject, I'm all ears. (no pun intended)

Re:By the time True AI is here...

[Anonymous Coward]

Absolutely. We may have different subjective opinions of, say, attractive or ugly, but at least we can verify that we both actually see the same thing. If Robotaster 2000 tells me the cookie is delicious, and it is in fact foul, I would begin to wonder whether there were commercial interests influencing the robotasters AI software. Cue the advent of a commerical-spewing bot that follows you round all day telling you what to buy.

Obvious?

[Bogtha]

it's also based on a very non-intuitive fact: they know what a sound was by knowing when there was no sound.

Surely there has to be more to it than that? Not only is it not "non-intuitive", it's completely bloody obvious, so much so that I already assumed that people did this in professional recording situations.

Think about it, it's just like weighing two things together, and then finding the weight of one item by weighing the other and taking the difference between the two measurements.

Re:Obvious?

[sidfaiwu]

I think that what the researchers did was very non-intuitive. They describe the sound mathematically by the locations of the silent points. This requires a much smaller representation than describing the multitude of locations that are non-zero. The assumption (I believe they use an actual theorem) is that the points of silence uniquely describe a sound. i.e. there are no two sounds that have the exact same set of zero-points. You are right in that subtractive synthesis of noise from a signal is a technique as old as signal analysis itself.

Re:Obvious?

[Anonymous Coward]

Obviously there must be more assumptions than that, since generally there exist lots of continuous functions having a given set of zeroes.

Obvious in retrospect

[Red Flayer]

Well, first off, FTA: "In fact, he notes, it may be the same type of method the brain actually uses. " (emphasis mine)

Second, we can all very easily deduce that our interpretation of sound deals with a very low signal-to-noise ratio. How many background sounds are we dealing with constantly? How surprising is it that analyzing sound subtraction (cancellation) from the noise is as effective as analyzing addition?

I'm no hearing expert, and I'm definitely not an expert when it comes to algorhythmic sound analysis -- props to the researchers who worked on this -- but wow, isn't that obvious in retrospect?

Re:Obvious in retrospect

[Elminst]

isn't that obvious in retrospect?

EVERYTHING is "obvious" in retrospect. The question always seems so simple once you know the answer.

Re:Obvious in retrospect

[bar-agent]

EVERYTHING is "obvious" in retrospect. The question always seems so simple once you know the answer.

42.

Re:Obvious in retrospect

[Elminst]

That does not apply. They didn't know the question.
In my statement, the article, and the parent post to mine; the question is already known. When you find the answer, the already known question doesn't seem so hard.

had to say it.....

[specific]

an algorithm for the 'sound of silence'......

would that be a "Simon & Garfunkle rhythm"?

This sounds like Zen processing

[Flying pig]

but let me be the first (and last) to ask "But will it be possible to run this Zen processing on top of Xen?"

Could noose for awl!

[Rob T Firefly]

May bee this will fie Nelly meme the and off half fast peach wreck ignition soft where.

Actually, this might really be useful

[zappepcs]

I've seen work (sorry no web links) where a guy did audio/acustic testing for the purposes of using audible sound to guide robots. Despite it not being intuitive, using standard PC sound hardware, he was able to get very accurate readings. Adding this algorythm to the mix of software he was using might actully give robots very useful audible sensors.

I'm not sure how it would work, but he was able to determine position and distance quite well, but was having some issue with the different densities of materials; say a heavy curtain clad window vs. a concrete wall etc.

By analyzing actual noise of the environment, and matching that to desired noises, it might be possible to use this to determine the accurate distances despite differences in target densities... and that would simply be very cool and useful for robotics. Not sure how that would map to underwater robots, but seems reasonable that it might help.

Sounds of Silence, Huh...

[Nom du Keyboard]

The Sounds of Silence. I'm sure that the original submitter, the Slashdot moderator, Slashdot itself, and everyone who downloaded the text time of "Sounds of Silence" when they opened this page will be hearing from the RIAA lawyers shortly. And for only $3000 in protection money, they'll leave you all alone until the next time.

Too late

[Hoi Polloi]

Sorry but Simon and Garfunkel already published their research on this.

Untitled

[FhnuZoag]

(n/t)

Original paper?

[moultano]

Half the time, I think people feel that science is out of their reach because the articles they read about it don't give them enough information to even start learning about it. When science is presented like this it gets reduced to blurbs with the logical content of zen koans. "sound from no sound" "noise from silence" A little more and I've got a haiku. I appreciate the need for simple summaries, but comeon, at least link to the meat of it.

Re:Original paper?

[iamlucky13]

Heck...a little bit more and you've got a Simon and Garfunkel song. Perhaps you could even explain why the words of the prophets are written on the subway wall. But I digress.

The summary is completely useless, and the article isn't much better. After reading the description 3 times, I figured out the graph, at least. X-axis is time, Y-axis is frequency, and color is amplitude, so it's essentially a time dependent power spectrum density (PSD) with anything above a cutoff amplitude shown in black. I believe the Navy uses a variation of this called a waterfall to help interpret sonar sounds. I got stuck again, though, reading their description of the sample.

The two researchers used white noise -- hissing similar to what you might hear on an un-tuned FM radio -- because it's the most complex sound available, with exactly the same amount of energy at all frequency levels.

Aside from the apparent infinite-energy contradiction if this were true, the graph clearly shows that the signal is both frequency and time dependendent. Obviously, that's the case they ultimately have to deal with to apply this method, but the article suggests otherwise.

As for what they're actually doing, presumably, instead of operating on a set of data that includes time, frequency, and amplitude, they are cutting it down to time, frequency, and sound/no sound. This would cut the data size by the amplitude resolution (eg, 1/16th for a 16 bit amplitude sampling). This must assume that amplitude is irrelevant to the sound, which based on my (limited) experience working with PSD's, I'm skeptical of. Perhaps the odds of getting the same digital time/frequency data for two different sources is low enough that this can be ignored, much like the likelihood of two data sets yielding the same MD5 sum is non-zero but insignificant.

Re:Original paper?

[swillden]

I appreciate the need for simple summaries, but comeon, at least link to the meat of it.

The second link in the slashdot summary includes a link to the full text of the research article.

Half the time, I think people feel that science is out of their reach because the articles they read about it don't give them enough information to even start learning about it. When science is presented like this it gets reduced to blurbs with the logical content of zen koans. "sound from no sound" "noise from silence" A little more and I've got a haiku.

I felt the same way, but in this case I'm not sure what else could be done. If you read the actual paper, what you find is that the key ideas in the paper are mathematical. The researchers chose to apply a different mathematical model (the reassignment class of time-frequency representations) to analyzing sound and discovered that it did a good job of isolating "interesting" components of sounds. This approach was motivated by consideration of the signal processing abilities and, more importantly, limitations of neurons, and refined with various other mathematical ideas. It appears to me (note that I don't actually understand the paper, so I could be wrong) that there really is no non-mathematical way to present these ideas, so any accessible presentation is going to be essentially content-free.

Giving people enough information to start learning about it in this case would seem to entail first giving them an overview of the mathematics of signal processing, including Fourier transforms, various time-frequency representations, statistical distributions of noise, etc. Perhaps once the researchers gain a deeper understanding of exactly why this modeling approach is so effective it will be possible to describe it in a simple analogy -- but it's also possible that the only way to understand it is through the math.

In this case, given the apparent impossibility of explaining this in a non-mathematical way, the author of the summary seems to have fixed on the researchers' observation: "Sparse representations are of great interest in neuroscience, particularly in auditory areas, because most neurons in the primary auditory cortex A1 are silent most of the time." Since the researchers' "reassignment" transform has the effect of producing distributions that are mostly zero or near-zero, the model seems to mimic the processing done by the auditory cortex, and provides a representation of the sound that suppresses sameness (to "silence") and highlights differences.

I don't think the author of the summary did a particularly good job with the explanation, but I can't really fault him for failing, because there just doesn't seem to be a good way to write it.

Re:Original paper?

[Taral]

Actually, there's no cutoff. They transform the standard sonogram using phase-derived data into what you see. It effectively concentrates information, lowering the entropy and clearly enhancing feature recognition. If you want to know more, read the paper (not the article -- the article is garbage).

Re:Original paper?

[john83]

Disclaimer: I'm a PhD student in pretty much the same field, but I'm not experienced enough that anyone should take my opinions as fact.

TFA looks interesting but overhyped to me.

The summary is completely useless, and the article isn't much better. After reading the description 3 times, I figured out the graph, at least. X-axis is time, Y-axis is frequency, and color is amplitude, so it's essentially a time dependent power spectrum density (PSD) with anything above a cutoff amplitude shown in black.

Not exactly. What they are doing is taking a normal time-frequency plot (as they say, like a music score) and transforming it to have new axes - the derivitive of phase and the group delay. They're arguing that this is generally a sparser representation of the signal than other representations. They analyse what white noise looks like in it. They give a hand waving physiological argument that this is how the ear might work, the strongest of which was the sparsity compared with the relative inactivity typical of the ear neurons - from TFA:

The fraction of the time-frequency plane occupied by the support of the distribution decreases as the sequence becomes longer, as in Fig. 3; therefore, reassigned distributions are sparse in the time-frequency plane. Sparse representations are of great interest in neuroscience (41-43), particularly in auditory areas, because most neurons in the primary auditory cortex A1 are silent most of the time (44-46).

The usual advantages of sparsity are in areas like compression, but with no inverse to their transform, I'm not sure that would be useful.

This is just another research paper guys, not a huge breakthrough or something.

Can it get the 18-1/2 minutes back?

[dbrower]

We're still waiting for an algorithm that will restore Nixon's "accidentally" erased Watergate tape. If ever there were a sound that wasn't there...

-dB

fantastic ?

[yet another fancy ni]

I couldn't find any factor of improvement so it's hard to say how fantastic this is and digging deeper i failed the math but wondered about if the beautiful pictures of white noise was just random number generators or equipment noise ? If anyone could tell me more I would be interested...

Not worthy of a press release

[the_povinator]

I work in the area of audio processing (speech recognition) and I can tell you that they
do not have anything worthy of a press release. They distinguished a pure tone from noise,
which is a very easy thing to do. Other than that, they just have a pretty picture. It's
not useful for anything.