The No.1 Website for Pro Audio
 All  This Thread  Reviews  Gear Database  Gear for sale     Latest  Trending
Hearing below the noise floor
Old 22nd November 2010
  #1
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb Hearing below the noise floor

In another thread someone suggested that he could hear soft artifacts even when they're 40 dB below the noise floor. That didn't make sense to me, so I created a series of test files, and added them as a Part 2 to my Artifact Audibility report. Here's a link to that part of the article:

Hearing Below the Noise

I'll be interested to learn if anyone here can honestly say they hear more than 20 dB below the noise floor in these test files.

--Ethan
Old 22nd November 2010
  #2
Lives for gear
 
Lute's Avatar
 

The tones-20.wav and tones-30.wav files are similar, but with the tones 20 and 30 dB below the noise respectively. I can just barely hear the tones in the -20 file, and I doubt anyone can hear them in the -30 version.

I could hear the tones in the -30 example.

I couldn't make out the tones too clearly when they were buried behind the pop music examples though.
Old 22nd November 2010
  #3
Lives for gear
 
Nordenstam's Avatar
 

Hi Ethan!

Here we go again.

How many nuances of cello intonation can you discern that normal audio people do not hear? I bet you could, if you wanted, create a bunch of examples of really small differences. Wee tiny stuff that you can hear and normal people do not grasp at a higher level of consciousness. It's not as much about physical ability, which is generally quite amazing, as it's about consciously noticing what the ears detect.

You know all too well there are people out there who consistently do blind test such things. The -30dB tone definitely needs no blind testing. It's not that hard to hear. The measurements also supports this as seen in this spectrogram of the tones-30.wav:


Generally speaking: if a noise floor can be heard, anything poking up above that noise floor can also be heard.

There's no clean file to blind test against, but I assure you that I can do more than 10/10 in a blind test. The lowest -50 music example was the only one I found hard to consciously hear. It's not possible to know how the gated tones affects the music without a clean reference for comparison. If you supply that, I'll have a go at blind testing.


An essential part to understand here is that there's so much more to sound levels than single deciBel numbers can convey! The -90dB figure of CD format doesn't really work that way as far as the ears are concerned. A -90dB sine wave is LOUD compared to the CD noise floor. The noise floor is spread out across the spectra. The hearing system uses about 24 frequency bands. What really matters is the energy within each of those bands. A flat TPDF CD noise floor measures about -113dB RMS energy in a 500 Hz band centered on 3kHz, loosely corresponding to a critical band in that frequency region. The least detectable change in each such band is about 1dB. Adding a -4dB signal to a 0dB signal results in an increase of about 1.25dB, which places it above the threshold for detection. (edit: oops! That should be 1.4dB. Still above threshold). This means that the dynamic range in the most critical range should be, loosely, 117dB for 16 bit TPDF dithering. As you know from Martins test, such dithering have been blind tested on normal loud music originally recorded on magnetic tape.. (more on those subjects in this and that post)


MBIT+ dithering have proved to much tougher to hear than TPDF or HPTPDF. It also have the maximum difference between peak signal level and hearable range. The 16 bit MBIT+ dither peaks at -78dBFS while the usable dynamic range in the most critical frequency region have a range of about 125 dB.


Using MBIT+ shaped dithering noise, blind testing the same tones as you used at -40dB was easy! This is how the signal looks in a spectrum analyzer:


Here's the same in normal FFT view:


Edit: this is MBIT+ dithering on "medium" setting, as can be seen later in this thread.

And this is how the ABX session went by:

foo_abx 1.3.3 report
foobar2000 v1.0.3
2010/11/22 20:35:43

File A: MBIT and beep minus 40.wav
File B: MBIT.wav

20:35:43 : Test started.
20:36:19 : 01/01 50.0%
20:36:38 : 02/02 25.0%
20:37:05 : 03/03 12.5%
20:37:18 : 04/04 6.3%
20:37:32 : 05/05 3.1%
20:37:48 : 06/06 1.6%
20:37:56 : 07/07 0.8%
20:38:05 : 08/08 0.4%
20:38:12 : 09/09 0.2%
20:38:23 : 10/10 0.1%
20:38:40 : 11/11 0.0%
20:38:55 : 12/12 0.0%
20:39:13 : 13/13 0.0%
20:56:31 : Test finished.

----------
Total: 13/13 (0.0%)

First attempt. No trying to "listen laterally", trying to gleam a more vague sense of quality instead of consciously hear specific artifacts, no mental tricks to make intuitive senses concrete. No "25/30" result, no doubt.

Next time, try to come up with a real challenge.

Hehe..

Thanks for making me do another blind test!



Best regards,

Andreas Nordenstam
Old 23rd November 2010
  #4
Gear Addict
 
RawDepth's Avatar
 

I feel that Ethan's challenge is somewhat subjective. Some ears may be more sensitive to 3000Hz than others. Some speakers (or monitoring environments) may be more sensitive to 3000Hz than others.

I could clearly hear the 3000Hz tones at -20dB but I could not detect any speech at that level. (Perhaps that was because the tones sound exactly like the beeper on my microwave oven.) At -30dB, I think I may have heard the tones, but am not so sure. If it were not fresh in my ears from a previous listen then I would never have noticed it in the -30 file.

Much may depend on the content, dynamics, and average RMS of the test signals in question. On the other hand, when I listen to a track or mix, I generally focus on the tonal qualities of the instrumentation. I usually don't focus on the background noise unless something is present enough to draw my attention.

Still an interesting thread though. Thanks Ethan.
Old 23rd November 2010
  #5
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by Lupo View Post
Generally speaking: if a noise floor can be heard, anything poking up above that noise floor can also be heard.
Sure, but we can also hear below the noise floor, which is the whole point of these examples.

Quote:
It's not possible to know how the gated tones affects the music without a clean reference for comparison. If you supply that, I'll have a go at blind testing.
I wasn't sure how many files to put on my site, but I have all the sources. Which do you want? Just the two tones mixed? Or the noise file alone? Or the music alone? Tell me which files you want and I'll upload them here for everyone to play with.

Quote:
A -90dB sine wave is LOUD compared to the CD noise floor.
Sure, but only if nothing else is playing and you crank the volume 40+ dB to be able to hear it. That's why I made the point in my article about a 3 KHz piercing tone "only" 10 dB below the noise is barely audible when the noise is at -40. But hearing the tones and noise at -90 while the music plays? Not in million years.

Quote:
Next time, try to come up with a real challenge.
Hey, where's your test? heh

This was just an attempt to bust the myth that anyone could possibly hear artifacts 40 dB below the noise floor, as was claimed.

--Ethan
Old 23rd November 2010
  #6
Lives for gear
 
sound_music's Avatar
 

lol, anyone that claims they can hear artifacts 40 dB below the noise floor (let alone anything else) should be in the dolphin wrangling business or something. forget about being an AE, you've missed the point!

there's no audio situation i can imagine where information that far below the noise floor would be useful?
Old 25th November 2010
  #7
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by sound_music View Post
there's no audio situation i can imagine where information that far below the noise floor would be useful?
This is a great point. Even if certain very soft artifacts or distortions etc were just barely audible, who cares? It's not like the entire musical experience is ruined. People who enjoy LP records put up with obvious clicks and pops and surface noise. And I can enjoy FM radio in my car even though the s/n ratio is mediocre at best. So why do so many people obsess over perceived "problems" that are 30 to 60 dB softer than those?

--Ethan
Old 26th November 2010
  #8
Lives for gear
 
filipv's Avatar
Quote:
Originally Posted by Ethan Winer View Post
This is a great point. Even if certain very soft artifacts or distortions etc were just barely audible, who cares?
Not me.

Quote:
Originally Posted by Ethan Winer View Post
So why do so many people obsess over perceived "problems" that are 30 to 60 dB softer than those?
They feel good about themselves that way?
Old 26th November 2010
  #9
Lives for gear
 
sound_music's Avatar
 

i think alot of guys make those claims and start those kind of debates because they're bored. killing time between sessions, not enough bookings, stuck at home with the flu, etc... IMO it's the kind of minutiae you could only enjoy if you had absolutely nothing else to do or talk about...

because hearing stuff down there has absolutely no practical value.
Old 26th November 2010
  #10
Lives for gear
 
Nordenstam's Avatar
 

Quote:
Originally Posted by Ethan Winer View Post
Sure, but we can also hear below the noise floor, which is the whole point of these examples.
Was writing "poking up above the noise floor" as in being detectable above noise floor in frequency domain analysis. As shown in the FFT and spectrum analyser pics above. A sine wave at -40dBFS "pokes up above" a 0dBFS noise floor with psychoacoustic noise shaping. In fact, it doesn't drown in the 0dBFS MBIT noise until the sine wave is close to -50dBFS. It was straight forward to blind test the -40dB sine vs the clean noise. With TPDF dithering(spectrally flat), the noise is discernible somewhere above -30dB. Pink noise is somewhere inbetween those.

So, if you want your article to make sense from a scientific viewpoint, you should inform that the -30dB signal should be detectable in the pink noise and that some signals can in some circumstances be detected 40 to 50 dB below the peak level of some noise shapes.


Quote:
Originally Posted by Ethan Winer View Post
I wasn't sure how many files to put on my site, but I have all the sources. Which do you want? Just the two tones mixed? Or the noise file alone? Or the music alone? Tell me which files you want and I'll upload them here for everyone to play with.
The clean files, without the beeps, so it can be ABX'ed against the dirty examples you created. (clean noise, clean music)

As you know, blind testing is the name of the game. The files you provided have no scientific value without a reference.


Quote:
Originally Posted by Ethan Winer View Post
Sure, but only if nothing else is playing and you crank the volume 40+ dB to be able to hear it. That's why I made the point in my article about a 3 KHz piercing tone "only" 10 dB below the noise is barely audible when the noise is at -40. But hearing the tones and noise at -90 while the music plays? Not in million years.
The established science says otherwise.

From the first file in JJ's powerpoint page on the AES site: PowerPoint Presentations from recent (or not so recent) meetings.

"As seen from the previous graph, the ear can hear to something below 0dB SPL at the ear canal resonance.
..
For presentation (NOT capture, certainly not processing), a range of 6dB SPL (flat noise floor) to 120dB (maximum you should hear, also maximum most systems can achieve) should be sufficient. This is about 19 bits."

Keep in mind that the 6dB SPL noise floor figure is for the entire spectra. The sensitivity in each of the hearings critical bands will be 10-15dB higher due to the resonances of the ears shape. Resulting in a hearing threshold curve that looks something like this:



This paper is also having a lot of good info on the subject of perception, check figure 7 and 20 for starters: http://www.meridian-audio.com/w_paper/Coding2.PDF


That's of course without music playing.

16 bit dithering noise and truncation artifacts are around -90dB peak level. This can be heard in various material. Truncation is especially easy to detect. Flat dither or high pass TPDF can also be blind tested in some circumstances. The only 16 bit dither that evades all blind tests, as far as I know, is psychoacoustically shaped noise. Which ironically peaks at a higher level than -90dBFS. As the MBIT+ dithering used in the example in previous post. This is so hard to blind test at 16 bit level because it provides a dynamic range of about 120dB in the most sensitive parts of the spectra. Whereas normal dithering that can be blind tested with full music running "only" have about 110dB dynamic range in the relevant spectral regions.



Remember the room noise measurement you posted from your own room?

This is your own noise floor relative to a 90dB SPL average/110dB SPL peak signal:


This is the same with a -100dB FS sine wave at 3kHz:


And this is the same as shown above truncated to 16 bit:

(from https://www.gearslutz.com/board/4907044-post278.html)

I see no reason to assume that keen sensitive listener should not hear that -100dB tone or the obvious truncation distortion when going to 16 bit without dithering. In your room, with your equipment.

I know you're claiming it'll all drown due to masking. As shown in Martins ABX test, this is not the case even with complex spectraly dense music peaking at -2dBFS.


Quote:
Originally Posted by Ethan Winer View Post
This was just an attempt to bust the myth that anyone could possibly hear artifacts 40 dB below the noise floor, as was claimed.
Well, it have been blind tested.. The myth you try to create is busted.

You set up a certain expectation bias, serve some files to confirm that expectation bias and lo and behold - the expectation bias does actually influence what people think and hear. That procedure is exactly the same as is all too common in audiophilia. And in good tradition with audiophilia, the results you achieve with such a test procedure does not mesh with established science.

If you decide to keep the article as is, ignoring science, pushing your own subjective world view onto other people.. Isn't that exactly what you so love to criticize audiophilia for doing?


Quote:
Originally Posted by sound_music View Post
i think alot of guys make those claims and start those kind of debates because they're bored. killing time between sessions, not enough bookings, stuck at home with the flu, etc... IMO it's the kind of minutiae you could only enjoy if you had absolutely nothing else to do or talk about...

because hearing stuff down there has absolutely no practical value.
You do it every day, consciously or not. It's why you're able to hear differences between various pieces of gear. Why you can detect that this plug in sounds good, that plug in doesn't.

The reason I care about such things is that I want my mastering chain to sound great. By technically testing everything and ensuring there are no abrassive artifacts present at potentially perceptible levels, the end result is nice clean sound. If I want it to be dirty, I want to be able to insert a good sounding noise or distortion. Having foul digital junk inserted in all projects seems like a bad idea to me.. I see no reason what so ever to have a processing chain that introduces small but potentially irritating artifacts into the audio signal.

Blind testing is for me, as for Paul Frindle and many others, the primary tool to assess the perception abilities. It's hard to argue with blind tests.

I don't obsess over this any more than I obsess over 600 or 500 milliseconds -60dB decay time of the standing waves in the room. One can always argue this or that is more important. Music isn't important compared to other matters. One billion people are hungry every day. This is about comparing apples to apples. If the choice is between a good apple, or a great tasting one, why not the great one?

This is just one of many aspects of technical side of audio engineering. With analogue processing, creating horribly bad sounding artifacts is next to impossible. With digital, it's impossible to avoid altogether. The only thing one can do with digital sound is to push the artifacts low enough. By doing technical tests on processing routines, it's easy to scrap those plugs that measure high levels of bad artifacts. There's little correlation with price in this. Some cheap, or even free, routines are as good as much more expensive options. Though it takes some searching to find those cheap routines.

The end result of keeping the processing chain clean is that I can have superb fidelity masters when that's needed. When the track calls for some gunk, I'll add some benign artifacts that actually makes it all sound better.


Best regards,

Andreas Nordenstam
Old 26th November 2010
  #11
Lives for gear
 
Nordenstam's Avatar
 

Quote:
Originally Posted by Lupo View Post
So, if you want your article to make sense from a scientific viewpoint, you should inform that the -30dB signal should be detectable in the pink noise and that some signals can in some circumstances be detected 40 to 50 dB below the peak level of some noise shapes.
Came to think about this. It was too quick and fast.

The truth is that a noise floor can be shaped to provide arbitary dynamic range within any bit depth. As shown in 1 bit high speed systems where the noise floor is technically about -6dB, yet the dynamic range in the essential area is (hopefully) around 110-120dB. With those systems, one is literally hearing well beyond 100dB lower than the peak level of the noise floor.


This one is more accurate:

"So, if you want your article to make sense from a scientific viewpoint, you should inform that the -30dB signal should be detectable in the pink noise and some signals can in some circumstances be detected 40 to 50 dB below the peak level of some noise shapes in 16/44 format. Beyond the 16/44 limit, noise shaping can be more extreme, providing arbitrary precision within the audiofrequent range even with a noise floor as high as -6dB."
Old 26th November 2010
  #12
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Thanks as always Andreas for your intelligent contributions.

Quote:
Originally Posted by Lupo View Post
A sine wave at -40dBFS "pokes up above" a 0dBFS noise floor with psychoacoustic noise shaping. In fact, it doesn't drown in the 0dBFS MBIT noise until the sine wave is close to -50dBFS.
Yes, but being able to see a signal in an FFT and being able to hear it are different because of masking, among other hearing foibles.

Quote:
So, if you want your article to make sense from a scientific viewpoint, you should inform that the -30dB signal should be detectable in the pink noise and that some signals can in some circumstances be detected 40 to 50 dB below the peak level of some noise shapes.
I'm an empirical kind of guy, and my main interest is creating practical tests that people can use to learn what matters with audio and what does not matter. I gladly leave the more scientific and scholarly work to people like you. I am serious.

Quote:
The clean files, without the beeps, so it can be ABX'ed against the dirty examples you created. (clean noise, clean music)
Attached are four files: Narration from my Analog Synth Programming video, the cheesy pop music, the combined tones alone, and the pink noise alone.

Quote:
blind testing is the name of the game. The files you provided have no scientific value without a reference.
I agree about blind testing, especially when settling a bet for real, but I think my files have a lot of practical value. heh

Quote:
If you decide to keep the article as is, ignoring science, pushing your own subjective world view onto other people.. Isn't that exactly what you so love to criticize audiophilia for doing?
Obviously I disagree. But now you have the source files, so I urge you to put them together at various mix levels and upload here, asking people to post where the tones start and stop in your versions. If a statistically significant number of people can identify where the tones start and stop when they're 30 or 40 dB below the pink noise, which itself is at a typical level -80 or -90 dB below the pop music, I'll be glad to concede the point. Until then, the burden is on you.

Quote:
You do it every day, consciously or not. It's why you're able to hear differences between various pieces of gear. Why you can detect that this plug in sounds good, that plug in doesn't.
As you know, I think most of that stuff is due to sighted bias. Which is why we need blind tests!

Quote:
The reason I care about such things is that I want my mastering chain to sound great.
Nobody disputes that, and I agree 110 percent that gear should be clean to below what is audible. Then you can run through several pieces of gear in series (or multiple generations) with no degradation. This is engineering 101, to over-specify a design up to the point where the cost is not unreasonable. But that's very different from some of the claims I see people make about their own hearing.

--Ethan
Attached Files

100 Hz + 3 KHz.wav (430.8 KB, 958 views)

Pink Noise.wav (866.0 KB, 791 views)

Romberg.wav (1.62 MB, 847 views)

Narration.wav (779.2 KB, 1161 views)

Old 26th November 2010
  #13
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by Ethan Winer View Post
If a statistically significant number of people can identify where the tones start and stop when they're 30 or 40 dB below the pink noise, which itself is at a typical level -80 or -90 dB below the pop music, I'll be glad to concede the point.
I would say that if any number of persons, even just one, can identify things a statistically significant number of times, you should concede the point.

Quote:
Until then, the burden is on you.
I entirely disagree that the burden is on Andreas. On the contrary, you are publishing an article about audibility, the burden rests on you to verify and prove every "fact" you present.

In the mean time:



foo_abx 1.3.4 report
foobar2000 v1.1
2010/11/26 18:28:20

File A: C:\Users\UnderTow\Desktop\Ethan\music_tones-40.wav
File B: C:\Users\UnderTow\Desktop\Ethan\romberg.wav

18:28:20 : Test started.
18:29:08 : 01/01 50.0%
18:29:43 : 02/02 25.0%
18:30:03 : 03/03 12.5%
18:30:16 : 04/04 6.3%
18:30:29 : 05/05 3.1%
18:30:39 : 06/06 1.6%
18:30:47 : 07/07 0.8%
18:31:19 : 08/08 0.4%
18:31:31 : 09/09 0.2%
18:31:55 : 10/10 0.1%
18:41:10 : Test finished.

----------
Total: 10/10 (0.1%)





foo_abx 1.3.4 report
foobar2000 v1.1
2010/11/26 18:43:34

File A: C:\Users\UnderTow\Desktop\Ethan\music_tones-50.wav
File B: C:\Users\UnderTow\Desktop\Ethan\romberg.wav

18:43:34 : Test started.
18:45:02 : 01/01 50.0%
18:45:27 : 02/02 25.0%
18:45:46 : 03/03 12.5%
18:46:08 : 04/04 6.3%
18:46:30 : 05/05 3.1%
18:46:51 : 06/06 1.6%
18:47:11 : 07/07 0.8%
18:47:31 : 08/08 0.4%
18:47:52 : 09/09 0.2%
18:48:09 : 10/10 0.1%
18:48:16 : Test finished.

----------
Total: 10/10 (0.1%)

As Andreas' result for the MBIT + tones at -40 test: First attempt. No "25/30" result, no doubt. I don't even have to listen to the entire clips or even both alternatives. I listen to part of A and can hear if there is a tone or not. I listen to part of X to hear if there is a tone or not. I know whether A is X or Y. In other words, it is obvious to my ears!

As Andreas said, Next time, try to come up with a real challenge. ;-)

In jest of course. It is always good to re-test one's assumptions and ideas.

I didn't test the noise files because you provided a mono noise file that has a different level to the files with the noise+tones. If you provide a stereo file at the same level as the noise+tone files (to make sure they are created in exactly the same way and that I am not getting a false positive from another difference than the tones), I will be happy to do some ABX testing on those too. (I can hear the tones when playing the files but we have to make sure I am not fooling myself).

You should really provide this stereo noise file in your article otherwise no one can properly test things.

Alistair
Old 26th November 2010
  #14
Lives for gear
 
sound_music's Avatar
 

Quote:
Originally Posted by UnderTow View Post
A much stranger thing I would say, is people posting in threads about topics that they say they don't care about and they think have no practical value. How weird is that?

You comments belittle all scientists and engineers worldwide in all and every field. Some of us consider ourselves audio engineers and care about the details and the science of our craft. Some of us even find it positively interesting!

....
woah slow down there Alistair, i think you're being a little touchy!

i didn't intend to belittle anyone; i was on a break killing time myself, (waiting for a singer to show up for overdubs), so maybe that's why i had that opinon at that moment. nothing more!

i have a huge respect for equipment designers and people who push the envelope sonically; i feel grateful for them every day. and i thought the OPs question was in relation to mixing; where my point stands BTW. in a mix context there's no useful info down there.

but i have to agree with you, the minutiae of this subject is not my cup of tea, i shouldn't have posted here--even if i was bored. again, no offence intended...

Old 26th November 2010
  #15
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by sound_music View Post
woah slow down there Alistair, i think you're being a little touchy!
Maybe I am. Sorry about that.

Quote:
no offence intended...
Great!

Alistair
Old 26th November 2010
  #16
Lives for gear
 
sound_music's Avatar
 

cheers
Old 26th November 2010
  #17
Gear Guru
 
Muser's Avatar
interesting test.

Thanks Ethan.

I found the only one very difficult to detect was the speech-20.wav. personally I believe we hear a lot more than we presume but we don't always detect it. these are two quite different things imo. I tend to lean towards Riemanns hypothesis that sound may be transmitted to the inner ear as a geometric whole or gestalt. it is then down to our use of perception, detection and pattern recognition which determines our ability to detect it.

I was listening to these on quite poor headphones and I believe I could detect the noises and tones. I would be very interested to listen on electrostatic headphones. however, the earlier examples are cueing me to detect the noise and tones and gestalt pattern recognition is understood to be at the cusp of the physiological and psychological. so it is often easy to think you hear something that isn't there (or) not be a result of the pattern you think you are detecting.

I have programmed such a large amount of synthesizer and sampler sounds that I am probably pretty well tuned into hearing noise and tones signals as part of an intelligible whole.

I guess I find 3 things significantly important.
one is the (hearing) which I take to be a physiological reality.
the second is (detecting) which I take to be more determined by experience.
the third is (identifying) which may be done by tools such as iZotope RX.

you may be able to identify an artifact with a tool like RX but what you hear or detect when the artifact is removed and your choice surrounding those decisions, I think is what ultimately makes a difference as regards good Audio editing.

I included relevant passages taken from

Analysis and Synthesis
On Scientific Method -
Based on a Study by Bernhard Riemann
REPRINT
Originally published in Systems Research, 1991, Vol. 8, No. 4, pp 21-41, Thesis Publishers, ISSN 0731
Revised version, 1996. (Downloaded from the Swedish Morphological Society


Hermann Helmholtz:
"One of the central features of Helmholtz' theory concerned the role and function of the
middle ear. Helmholtz contended that when a complex tone is transmitted from the
eardrum to the cochlea (inner ear), by way of the ossicles (the three small bones in the
middle ear), it is transmitted not as a geometrical whole, but in the form of a set of primary
tones. The middle ear "breaks down" the complex tone into its primary components, each
component consisting of a given frequency and intensity. Each primary tone then
stimulates a specific area or resonator in the cochlea, and each resonator then transmits
a signal to the brain by way of a specific nerve. This is what Helmholtz called his "theory
of specific energies".


Bernhard Riemann:
"Riemann's manuscript contains a number of important
conclusions. Firstly, he rejects Helmholtz' theory that the middle ear only transmits
primary tones to the cochlea. To the contrary, the ear must "transmit to the fluid of the
inner ear the variation in the air pressure at every moment at a constant ratio of
amplification. ... we consider it justified to assume that the timbre curve is altered only
very slightly by the transmission".

Riemann's ear perceives the "quality of sound" -- its timbre curve -- as a geometric whole
or gestalt. This also means that Riemann rejects Helmholtz' ad hoc hypothesis that the
middle ear "distorts" the sound wave in order to produce harmonics and combination
tones.

Finally, Riemann applies the principle of "projective invariance" as concerns the
transmission of sound waves in the ear. He argues that some physical quantity inherent in
the geometry of the wave pattern must remain invariant in its transmission from the
eardrum to the cochlea -- if we are to explain what the ear actually accomplishes. He
identified timbre -- a term that is usually referred to as the "quality of sound", but which is
quantifiable -- as such an invariant.

For Riemann, one of the more intriguing problems to be understood is the ear's ability to
perceive sound waves, the energy levels and physical displacements of which are so
small that they cannot even be measured directly. Modern quantitative measurements
of the human ear's sensitivity reveal that the sensory receptors of the cochlea can detect
motions of atomic dimensions; and that we can perceive musical tones in which the
energy transmission to the eardrum is of the order of 10-18 Joules."

Last edited by Muser; 27th November 2010 at 12:54 PM.. Reason: small alterations
Old 26th November 2010
  #18
Gear Guru
 
UnderTow's Avatar
Oh btw Ethan, unless I overlooked it by accident, I can't see the concerto sample without any noise burst. (The original). I would like to ABX test the audibility of the noise burst there too. Can you upload it?

Thanks,

Alistair
Old 27th November 2010
  #19
Lives for gear
 
Nordenstam's Avatar
 

Great to see you're such a sport, Ethan!

Coming back to the specifics after a diversion to philosophers corner..

Physics, measurements and blind tests are all good - but I don't think that's what this really is about. Don't think we can blame practical hearing abilities. They are generally very good for anyone who sets out to hear some particular aspect in the endless world of sound.

And there's the point. Interpretation of sound is a learned ability.

Notes and composition are literal languages. Recognizing different instruments and particular ways to treat them is also a word driven process. That's not what I have in mind though. It takes more than note reading skills to be a good sound artist, in the widest sense. What can't be said in words needs, like the literal languages, to be learned/trained to recognize in order to be deliberately useful. Deliberate, yet not conscious in the sense "+1dB @100Hz q=.7" is. Many sides of hearing are better done unconsciously. Let me stress again that it does not mean without aim!

We learn to recognize certain abstract aspects of sounds like we learn the more easily describable stuff like name of notes, instruments and effects like compression. Listening to the rest, whatever you want to call it, is a sort of listening that lends itself particularly well to intuitive work.

Think that's the primary reason many feel blind tests to be artificial compared to normal listening. It's enforcing the square world of words and truth tables onto the wordless language.

Have learned to somewhat successfully tune out/in when doing blind testing. The ABX results from listening consciously to this or that aspect of sound is consistently about 10-15 dB worse than when I do the "lateral listening".

And this is purely a belief: that most normal people listen to music that lateral way. Intuitively, directly. Having a conscious relation to sound doesn't necessarily make for a better listener.

It obviously have to be detected by the consciousness in some way to be blind tested. Hope you can understand what I mean by wordless percept. Languages without words.

A person can be fluent and able to go to sublime depths in some of those languages, yet not being capable of understanding a single word in a totally foreign language. I bet anyone who hears those audio examples are perfectly capable of discerning way smaller differences in their native sound languages. Even if they happen to not hear the difference between mp3 or wav, dither or not, tape or digital recording, the wife lying or being honest, this or that.. No matter what sort of sounds one choose for such a test, it's nearly guaranteed to be far off from the particular individual listeners own specialty! If the test is to be relevant, it have to ask the question in the listeners own native language.

For musicians, that would typically be their favourite instrument. Audio engineers would typically be able to pinpoint weird details no one else cares about. heh For most people it would probably not have anything to do with consciousness. Many have little need for sound recognition beyond spoken language detection. That doesn't mean they're deaf. Intonations in voices can be dissected to an amazing detail. People do hear all sorts of minutae, even if they do not think about it. Have been surprised many times by the amazing ability people have to comment on details I struggle to hear myself. And anyone with normal ears are capable of extreme detail recognition if pressed in a crisis. The hearing system is an insanely deep sensory apparatus. It must mean that evolution have favoured the keen eared! I'm certain that the general potential for extreme acuity is there for all normal trios of brain-ears. Am equally certain that the result of not hearing those differences is the function of higher level filtering. As witnessed by the positive blind tests. If some can, just about anyone can given the right incentive and determination.


.. Trying to put some intuitions into words.. Is it smart to post in the middle of the night?
Old 27th November 2010
  #20
Gear Guru
 
Muser's Avatar
actually Ethan. I listened to the ones on your site. hope I was listening to the right ones. one thing which stood out was the added noise on the strings, made the natural bow/string scrape sound a little unnatural.
Old 27th November 2010
  #21
Lives for gear
 
Alexey Lukin's Avatar
 

Quote:
Originally Posted by Lupo View Post
Generally speaking: if a noise floor can be heard, anything poking up above that noise floor can also be heard.
Andreas, this is not quite accurate. The frequency resolution of FFT can be made arbitrarily high (by increasing FFT size), and it will allow you seeing tones of arbitrarily low level buried in noise (-60, -100 dB below noise, and so on). However the frequency resolution of our ear (and more importantly — its energy integration properties) are limited. So, FFT can (in this situation) show you more than we can actually hear.

You and Ethan started from a confusing concept of hearing below the noise floor. The source of confusion comes from the fact that noise spectrum can be colored, and the test tone may reside in "easy to detect" or "hard to detect" frequency ranges.

I think that from a psychoacoustical point of view, detectability of a tone depends on a ratio between the tone and the noise energy in a critical band (not equal and not even related to overall energy of the noise).

Quote:
Originally Posted by Muser View Post
Hermann Helmholtz:
"One of the central features of Helmholtz' theory concerned the role and function of the middle ear. Helmholtz contended that when a complex tone is transmitted from the eardrum to the cochlea (inner ear), by way of the ossicles (the three small bones in the middle ear), it is transmitted not as a geometrical whole, but in the form of a set of primary tones. The middle ear "breaks down" the complex tone into its primary components, each
component consisting of a given frequency and intensity."

Bernhard Riemann:
"Riemann's manuscript contains a number of important conclusions. Firstly, he rejects Helmholtz' theory that the middle ear only transmits primary tones to the cochlea."
I believe that the current understanding is that the frequency decomposition is performed by the cochlea, not prior to it (because cochlea has a place-variable stiffness, hence place-variable resonant frequencies). And the frequency selectivity of this decomposition (described by critical bands) is the primary factor determining detectability and masking.
Old 27th November 2010
  #22
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by UnderTow View Post
I would say that if any number of persons, even just one, can identify things a statistically significant number of times, you should concede the point.
Yes, agreed. If even one person can hear where the tones start and stop when they're 30 or 40 dB below the noise floor, and do that enough times, that proves that some people can hear 30 or 40 dB below the noise floor.

Quote:
I entirely disagree that the burden is on Andreas. On the contrary, you are publishing an article about audibility, the burden rests on you to verify and prove every "fact" you present.
I did prove that nobody can hear signals 40 dB below the noise floor with my test files. For example:

Quote:
foo_abx 1.3.4 report
foobar2000 v1.1
2010/11/26 18:28:20

File A: C:\Users\UnderTow\Desktop\Ethan\music_tones-40.wav
File B: C:\Users\UnderTow\Desktop\Ethan\romberg.wav
Great, you showed that you can hear the tones when they are at the same level as the noise floor. My article even says the tones are audible at that level. Now prove that you can hear the tones and speech when they're 30 dB below the noise floor, and then we'll have the basis for a discussion. heh

Is the intent of my test files really that unclear? I've explained it at least three or four times so far in this thread, beside the clear explanation in the original article. Here it is again: The original claim was being able to hear artifacts that are 40 dB below the noise floor of a digital recording. I don't recall if that was 16 bits or 24 bits, so let's give the benefit of the doubt and say 16 bits. If nobody can hear my tones when they're 30 dB below a -40 noise floor, that busts the myth. But if someone can hear the tones 30 dB below -40, then I'll take the next step and make files with tones 40 dB below -60 or -70.

Quote:
you provided a mono noise file that has a different level to the files with the noise+tones.
I loaded that mono noise file into SONAR and left the pan-pot in the middle. So I don't understand what you are unable to duplicate.

--Ethan
Old 27th November 2010
  #23
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by Muser View Post
I tend to lean towards Riemanns hypothesis that sound may be transmitted to the inner ear as a geometric whole or gestalt. it is then down to our use of perception, detection and pattern recognition which determines our ability to detect it.
Again, I'm more of a big-picture, practical type of person. I'm not interested in how or why as much as if. If nobody can say "Yes, there it is" then I'm satisfied to conclude that it doesn't matter.

As for my concerto, it's over 16 minutes long, and the original Wave file is 170 MB. So it's not practical to upload.

--Ethan
Old 27th November 2010
  #24
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by Lupo View Post
Great to see you're such a sport, Ethan!

<snip>

Trying to put some intuitions into words.. Is it smart to post in the middle of the night?
Well, at least you're not still claiming to hear artifacts 40 dB below a -90 noise floor. heh

Thanks Andreas, you da' man.

--Ethan
Old 27th November 2010
  #25
Registered User
 

Exclamation

Quote:
Originally Posted by Ethan Winer View Post
i dont hear noise in concerto-70
but i do hear in concerto-60.

usign laptop speakers.
pavillion dv7


i think i hear tones at -30, but i start to doubt if its my imagination.
i hear at tones at -20, no doubt.

i hear music_tones-40, no doubt.
i hear also music_tones-50, but... not as clear.

i hear "demo" noise in men_at_work2-40
also in the second part of men_at_work1-40. the first part its too bright.

i hear speech-20 but not as clear. just blabla bla.
i hear speech-10 no doubt.

i think to my self:
why is this so hard to listen??

then i remember: i havent turned off the TV & close the windows!!
LOL
its raining, and commercial air/jet planes are passing by..., my right ear its filled with wax, english ios not my native lenguage.
but thats no excuse...
Old 27th November 2010
  #26
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by Ethan Winer View Post
Great, you showed that you can hear the tones when they are at the same level as the noise floor.
Pay attention:

File A: C:\Users\UnderTow\Desktop\Ethan\music_tones-50.wav
File B: C:\Users\UnderTow\Desktop\Ethan\romberg.wav

Quote:
I loaded that mono noise file into SONAR and left the pan-pot in the middle. So I don't understand what you are unable to duplicate.
How are we supposed to know how you created these files? What pan laws are you using? Did you dither and if so with which dither? Etc etc etc.

I checked and that Pink Noise.wav peaks at -8,08 dB and has an average RMS of -19,96 dB while the tones-30.wav file peaks at -6,1 dB and has an average RMS of -16, 03 dB RMS. So I stopped there. We can't generate the clean Concerto file either.

Ethan, can you provide all the right reference files?

Alistair
Old 27th November 2010
  #27
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by Ethan Winer View Post
As for my concerto, it's over 16 minutes long, and the original Wave file is 170 MB. So it's not practical to upload.
We only need the exact same section you used to add the tones to of course...

Alistair
Old 29th November 2010
  #28
Gear Guru
 
Ethan Winer's Avatar
 

Lightbulb

Quote:
Originally Posted by UnderTow View Post
Pay attention:
LOL, you got me, I missed your second screen cap. But still, those tones are then 10 dB below the -40 noise floor, not 40 dB below which is the point of this thread.

Quote:
How are we supposed to know how you created these files? What pan laws are you using? Did you dither and if so with which dither? Etc etc etc.
As I (think I) said earlier, I just eye-balled the initial noise level and mentally averaged it. The subsequent -dB changes were then set accurately using SONAR's volume controls. No dither, noise panned center. Here's the Pan Law setting I use in SONAR:

0dB center, sin/cos taper, constant power

Quote:
can you provide all the right reference files?
I think you have everything except for the Concerto excerpt. That file is almost 14 MB so I put it on my web site for people to download. I won't leave it there forever, but I'll keep it for at least a week. Here's the link:

Concerto Wave file (14 MB)

If there are any other files I'm overlooking, let me know.

--Ethan
Old 30th November 2010
  #29
Lives for gear
 
Nordenstam's Avatar
 

Ethan, read post number 3 again. The blind test was tones peaking at 40dB below the peak level of the noise.

Please also have a re-read of Frindles post on the subject. It was a very special case of partially broken dither code. The audibility issue disappeared when the dithering issue was fixed. Such a scenario can not be recreated using normal dithering routines.
Old 30th November 2010
  #30
Lives for gear
 
Nordenstam's Avatar
 

Quote:
Originally Posted by Alexey Lukin View Post
Andreas, this is not quite accurate. The frequency resolution of FFT can be made arbitrarily high (by increasing FFT size), and it will allow you seeing tones of arbitrarily low level buried in noise (-60, -100 dB below noise, and so on). However the frequency resolution of our ear (and more importantly — its energy integration properties) are limited. So, FFT can (in this situation) show you more than we can actually hear.

You and Ethan started from a confusing concept of hearing below the noise floor. The source of confusion comes from the fact that noise spectrum can be colored, and the test tone may reside in "easy to detect" or "hard to detect" frequency ranges.

I think that from a psychoacoustical point of view, detectability of a tone depends on a ratio between the tone and the noise energy in a critical band (not equal and not even related to overall energy of the noise).
We're entirely on the same track! The critical band description was included some paragraphs below in my post. (although that's about as a vaguely related to reality as the paragraph you did quote :( )

It was meant as a very coarse hint. In this situation, it's easy to detect with measurement gear. When it is, and the levels in questions are firmly within hearable range, it should be a strong hint to keep listening! While it is technically possible to measure things that way that can not be heard, it's much more likely that such obvious differences can be heard. As seen in the blind tests.


Quote:
Originally Posted by Ethan Winer View Post
Well, at least you're not still claiming to hear artifacts 40 dB below a -90 noise floor. heh
Huh? I've blind tested -40dBFS vs 0dBFS. Moving that to -130 and -90 will give the same results.

Here's a direct link in case you don't bother to look through the relevant thread again: Paul Frindle - Is This Truth Or Myth? - - notice that the artifacts where around -110dB and it was claimed that those where about 40dB below the noise floor of the tape recording originally being digitized. It was also a very special case of correlation messing with the stereo field. It have no relation to the tests you made neither does it relate to 16 or 24 bit noise floors.


Let me also remind you of this ancient thread: Which amount of jitter makes the sound worse

Post #126: blind tested -120dBFS 2.5k sine wave in the blast of -90dBFS 16 bit TPDF noise floor. On a drum sound sample peaking at -16dBFS with an RMS power of -21dBFS. My typical RMS tracking level. 30dB into -90dB noise.
Post #128: blind tested -125dBFS 2.5k sine wave in 16 bit MBIT+ noise floor. The peak level of that noise is -78 dB, 47 dB above the -125dB sine.

The sample used in those tests is not a full scale sound. It is however at typical tracking level. It also happens to be an average level that often enough is used for end user delivery in "audiophile" music. The sort of music that is most likely to be played back on revealing systems for scrutinizing ears.

It's basically the same conclusion as the papers linked to earlier. The perfect 19-20 bit / 110-120dB performance demands a spurious free dynamic range of about 140-150dB as seen in the typical views of Fourier space.


Quote:
Originally Posted by Ethan Winer View Post
Thanks Andreas, you da' man.
A big thanks back at ya! We don't always agree, but it's always a pleasure to discuss with you.


Cheers,

Andreas
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump