• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Can we hear the bottom bits of 24?

OP
E

earlevel

Addicted to Fun and Learning
Joined
Nov 18, 2020
Messages
545
Likes
776
First, set a listening level for playback of the loudest parts.

For my convenience, let's pretend it is as loud as someone shouting at you from one meter. Too loud, but, it circles back to something I wrote long ago.

Assuming a set of theoretically perfect listening conditions (that don't exist), the distance of a second shouter's shout reaching you at the level of the last little bit of 24 would be 10,425 miles away...

If "the last couple of bits" is of interest, move that shouter to 5,212 miles distant.

Thanks, I've looked at that post before, I'll have to remember to use it when I have trouble relating these levels to someone ;)

A lot of audio engineers feel they can hear it anyway, might not make a difference, but maybe it will give some pause...
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,204
Likes
16,986
Location
Riverview FL
Thanks, I've looked at that post before, I'll have to remember to use it when I have trouble relating these levels to someone ;)

A lot of audio engineers feel they can hear it anyway, might not make a difference, but maybe it will give some pause...

My experience says, that starting with a loudish listening level, -72dB (lowest 4 bits of 16, lowest 12 bits of 24) or so is at the edge of inaudibiliy.
 

Head_Unit

Major Contributor
Forum Donor
Joined
Aug 27, 2018
Messages
1,341
Likes
688
Let's not forget that you can detect tones 20db below the noise floor. Do with that as you wish.
That's a common statement that I don't think is really true. It came up in discussions of aircraft overflight noise, as an issue the noise-uncaring FAA was trying to ignore in order to restrict overflights as little as possible (along with your hearing's directional ability and ability to detect patterns like helicopter rotor chop).
- Typically "noise floor" is a number, like -90 dB on an FFT. But maybe each tone adding up to that -90 is down at let's say -120. Now if there is a tone at -100, it is louder than the -120 tones next to it, and you can hear it. If it was next to -90 dB tones, they would drown it out. So we might hear tones that are below the total noise-that is where the phrase "listening into the noise" arises I believe. But we are not really hearing below the noise floor I don't think.
- Now what if you have a 60 dB or whatever tone from a speaker in front of you, then a similar frequency tone at 50 dB behind you, can you hear the one behind you? This is a question I don't know and would like to know the answer.
 

audio2design

Major Contributor
Joined
Nov 29, 2020
Messages
1,769
Likes
1,830
That's a common statement that I don't think is really true. It came up in discussions of aircraft overflight noise, as an issue the noise-uncaring FAA was trying to ignore in order to restrict overflights as little as possible (along with your hearing's directional ability and ability to detect patterns like helicopter rotor chop).
- Typically "noise floor" is a number, like -90 dB on an FFT. But maybe each tone adding up to that -90 is down at let's say -120. Now if there is a tone at -100, it is louder than the -120 tones next to it, and you can hear it. If it was next to -90 dB tones, they would drown it out. So we might hear tones that are below the total noise-that is where the phrase "listening into the noise" arises I believe. But we are not really hearing below the noise floor I don't think.
- Now what if you have a 60 dB or whatever tone from a speaker in front of you, then a similar frequency tone at 50 dB behind you, can you hear the one behind you? This is a question I don't know and would like to know the answer.

You not thinking it is true, does not make you correct. Read my statement. It is totally factual and easily verified.
 

AudioStudies

Addicted to Fun and Learning
Joined
May 3, 2020
Messages
718
Likes
398
Hear is some measurements I made of a sigma delta DAC showing its 23rd and 24th bit output levels.

Was "hear" an intentional pun?
 
OP
E

earlevel

Addicted to Fun and Learning
Joined
Nov 18, 2020
Messages
545
Likes
776
You not thinking it is true, does not make you correct. Read my statement. It is totally factual and easily verified.
But I think you'd agree that it's incomplete ("Let's not forget that you can detect tones 20db below the noise floor"). True when the tones are easily identifiable (easy to hear frequencies), the noise floor itself is loud enough (if the noise floor is below or near the edge of hearing, you're not going to hear tones yet another -20 dB), and it doesn't say you'll hear the tones if audio content much louder is going on.

In other words, it seems to be aimed at the special case of either a high noise floor, or one turned up to the point of being easily heard, with no other signal than the one you're trying to detect. No doubt that people are built to pick out sounds that are lower than a nearby waterfall, for instance. But that's different from having a noise floor that can barely be heard, or barely not heard, and expecting to hear something 20 dB below it.

I see the "20dB below the noise floor" claim most often when people are championing noise shaped dither, when the noise has been moved out of the way of the signal, of of course that's trivially provable. But when when people talk about using noise shaping on 24-bit truncations, it's just absurd. Even for 16-bit, talking about noise shaping letting a listener hear to -110 dBFS is still a sketchy proposition. Which is to say, I doubt the musical proposition of it. (It gets funnier when someone invents a new noise shaping they think will eke out another dB or fraction thereof of "perceptual" improvement...for something this is almost certainly imperceptible except under unnatural conditions like no music, along with gain that would be impractical for listening to music.) :)
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
The reason you can hear tones below the noise floor is the same as why enlarging an FFT to more bins lowers the noise in each bin. Our ear has 30 or so critical band filters. Since each filter is less than the total audible range the noise it responds to in that band is lower than the total noise over the audible bandwidth. It all comes together so that we can hear below the overall noise floor the deepest around that 3-5 khz range where our hearing is most sensitive. You can hear 15 or maybe 20 db below the basic noise levels here. Noise typically is pinkish to brownish anyway.

So maybe the noise floor in your listening room is 35 db SPL. In that 3-5 khz range it will be lower as most of the noise is below 200 hz. In the 3-5 khz range where our critical band filter is about 1/3 of an octave wide the noise we'd hear is lower than the total noise floor. In those most sensitive frequency ranges in room noise is likely only 10 db SPL or so. If there is nothing else going on, we can hear below that level so 0 db SPL can likely be reached by our ears in that situation.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
622 Hz is below our most sensitive hearing range, yet near the perceptual/octave center of the 20 - 20,000 range. I made a little experiment and found that -114 dB is easily audible in dithered 16-bit audio. That's about 20 dB below the dithered 16-bit noise level.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
622 Hz is below our most sensitive hearing range, yet near the perceptual/octave center of the 20 - 20,000 range. I made a little experiment and found that -114 dB is easily audible in dithered 16-bit audio. That's about 20 dB below the dithered 16-bit noise level.
You have to be careful using just a tone like that I think. Maybe your speaker has a several db peak in response right at that tone of 622 hz. But in any case your experiment shows you can hear below the basic noise floor.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
You have to be careful using just a tone like that I think. Maybe your speaker has a several db peak in response right at that tone of 622 hz. But in any case your experiment shows you can hear below the basic noise floor.
Yes, well below in fact. That tone at -20 dB below the noise floor is so obviously easy to hear, I speculate the perceptual floor is at least -30 dB or more. Of course, this is an ideal case, a single pure tone that is easy to differentiate.
 

audio2design

Major Contributor
Joined
Nov 29, 2020
Messages
1,769
Likes
1,830
622 Hz is below our most sensitive hearing range, yet near the perceptual/octave center of the 20 - 20,000 range. I made a little experiment and found that -114 dB is easily audible in dithered 16-bit audio. That's about 20 dB below the dithered 16-bit noise level.

At what gain though? i.e. how many db (sound) would a full scale signal be? ... however, this was sort of my point anyway, hence why I left the statement unbounded.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
I was only measuring it in relative terms: how far below broadband noise we can detect an embedded music or a non-random signal?
It may be that this relative difference/ratio also depends on the absolute total level of the sound. I did not measure that.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
Here's a wide dynamic range clip you can try. I created it by modifying a track on a CD I bought in the 1980s. It was a dynamic range demonstration. I modified it to increase the dynamic range further. This is a modified partial track sound clip for educational use.

The loudest part is at 3:00 (-1.37 dB digital scale). Set the volume to the loudest you or your system can take at that point. The loud part is jet noise which is broad spectrum, so an SPL meter should give you an accurate reading of how loud it is. Then leave the volume there and play it from the start. At what point can you start to hear anything?

NOTE: don't just crank the volume to max, that defeats the purpose, of course you can hear it all that way but the loudest part will blow your system or your ears. The point is to set the volume so your system & ears can handle the loudest point at 3:00, then see how far down you can hear from that point.

Here are approximate dB levels by time:
0:10 -90
0:20 -80
0:30 -75
1:00 -70
2:00 -60
2:50 -60, volume ramps up
3:00 -1.2

Doing this in my listening room, with the volume set as above, I start to hear sound at 15-20 seconds in. This means I have about 80 dB of perceptual dynamic range in my listening room. But that's much too loud for music listening, I turn the volume down 12-24 dB from this, so I've got about 60 dB of practical musical dynamic range. I live in a quiet neighborhood and my listening room is in my basement, ideal conditions.

With a slightly noisier system or room, the first thing you hear might be the birds chirping around 25 secs in. That is about -75 dB. If the first thing you hear is the water around 1:30, that is about -65 dB.
 
Last edited:

Head_Unit

Major Contributor
Forum Donor
Joined
Aug 27, 2018
Messages
1,341
Likes
688
You not thinking it is true, does not make you correct. Read my statement. It is totally factual and easily verified.
In my own mind it does ha ha. How are you defining "noise floor"?

What I meant was, looking at this thread for instance https://www.audiosciencereview.com/forum/index.php?threads/smsl-su-6-review-desktop-dac.28402/ and the graph right below the text "USB Jitter and noise shows some negligible (and utterly inaudible) unwanted tones:"
- Above in that post the "noise floor" is measured as -120 dB. Presume that if you added all the FFT bins in the graph they sum to -120. I believe you are asserting that you could hear tones at -140 dB. Assuming we amplify that into audibility, I'd agree, since the tones are still 15 dB above what I would call the "noise floor" of all the tones at around -155.
- What I'm saying is I don't believe single tones at -155 or below are audible. I don't recall ever seeing evidence you can hear tones at a significantly lower level than nearby tones. Somewhat lower perhaps. I'd like pointers to any research about it; my friend could use such.
- And yeah the numbers shift around depending on how many FFT bins blah blah blah
In other words, summing the noise floor to a single number, then saying we can hear tones "lower than the noise" I think can misdirect into a wrong impression.
 

Frgirard

Major Contributor
Joined
Apr 2, 2021
Messages
1,737
Likes
1,040
The cachalot reach 230 dB. So 24 bit is not enough.
My dream a duo by miss Cachalot and sister Atlas Rocket.
 

Somafunk

Major Contributor
Joined
Mar 1, 2021
Messages
1,311
Likes
3,028
Location
Scotland
Good informative video by Sonicscoop : Bad arguments for high resolution audio

 

Somafunk

Major Contributor
Joined
Mar 1, 2021
Messages
1,311
Likes
3,028
Location
Scotland
Not watched that one, cheers
 
Top Bottom