High Resolution Audio?

[GrimSurfer]

According to this page it's just A-weighting, slow integration. Nothing fancy, it's just the total noise level with a basic perceptual weighting, just like you'd see on a spec sheet for example. Which makes sense, since that will give the highest number anyway (larger, by definition, than any tone in the signal), and when you're dealing with safety standards you want to err on the safe side. Using that method will likely also give you around 30 dBA in a quiet room, I'd assume.

Exactly.

My listening space (living room, quiet suburban street, leafy neighbourhood, interior walls treated with Roxul Safe 'n Sound, R16 outer walls and R24 ceiling, medium weight curtains covering double pane argon-filled windows, a sealed fireplace, and an insulated and sealed door) measures 30 dB in the evening.

It's 29 dB in the winter with 3-6 feet/1-2 M of snow on the ground and roof (snow being a really good attenuator) outside. That's without the HVAC running.

If I tried really, really hard I might be able to knock this down by 1-2 dB.

So 10 dB for a normal residential space is simply absurd. Even typing it stretches the imagination.

 

[eliash]

See @edechamps very pertinent posts 205 and 208.

Thanks for the reminder, this brought me a lot further in understanding the issue, but to be honest, at that point I did not realise that the broadband ambient noise sums up that high, in contrast to the human "build in" critical band analyser, which yields much better SNR thresholds...

 

[solderdude]

Since we hear in Phon (not dB SPL) we might measure 30dB (not dBA) in a quiet room but in the part of the freq, range where we hear actual noise as 'noise' it looks like Ray's measurements show 10dB noise floor isn't that far fetched.
When you overlay Phon curves with the hearing limits Ray's room measures about 10 Phon ?
It basicaly means the room is quiet.

Then it is possible to hear faint tones above that noise floor in the 1-4kHz range thereby concluding that you can hear that low and need a certain dB range.
The trouble is, in practice, music isn't a continuous tone but an addition of many tones spread over a wide band where most of them will always be below 0dBFS so these will be inaudible anyway when peaks (not peaks of individual tones) register at 0dBFS.

To illustrate this I created a file which will show you that 80dB attenuation of music is enough to not hear anything anymore.

 

[eliash]

Exactly.

My listening space (living room, quiet suburban street, leafy neighbourhood, interior walls treated with Roxul Safe 'n Sound, R16 outer walls and R24 ceiling, medium weight curtains covering double pane argon-filled windows, a sealed fireplace, and an insulated and sealed door) measures 30 dB in the evening.

It's 29 dB in the winter with 3-6 feet/1-2 M of snow on the ground and roof (snow being a really good attenuator) outside. That's without the HVAC running.

If I tried really, really hard I might be able to knock this down by 1-2 dB.

So 10 dB for a normal residential space is simply absurd. Even typing it stretches the imagination.

Thanks for repeating the measurement, I really appreciate that, btw. it did´nt become clear to me which was the faintest level you could hear in your listening position?

Regarding the "absurd 10dB" I can follow you now. With that type of broadband measurement (20-20KHz for regular HiFi use, or 31-8KHz for professional SPL meters, so around 3dB less, because it´s less than half of the measurement bandwidth) you will end up even with higher values for the ambient noise.

The issue though is another one. The ear reacts to that broadband SPL in a different way, namely in those critical bands (or ERBs or MELs as alternative definitions) given e. g. here:
https://de.wikipedia.org/wiki/Frequenzgruppe
It is in German but the table should be clear. Here you have bandwidths shown at 100Hz upwards. As far as I understood human hearing, these frequency bands get detected individually by the inner ear and the "brain postprocessing". That means for your 441Hz tone, it would be detected in Band 5 ranging from 400-510Hz with a measuring bandwidth of 110Hz. So that cuts out most of the ambient broadband SPL. In case of a 20-20K measurement, only a fraction of 1/180 of the noise bandwidth is left. So the ambient noise level in that range should be down by a square root of 180 which is some 23dB less (assuming a constant background noise over frequency). If this is true for all the other bands, your personal hearing capabilities range far below the ambient broadband SPL. That´s why I could hear that tone at -96dB.

Here I have to correct myself also in the original SPL calcuation. I stated 4dB for the increased hearing distance of 2.5m compared to the speakers SPL spec at 1m distance. It must be 8dB, because the distance ratio is squared (did a multiplication of 10 instread of 20 in the dB calculation). Together with a really measured speaker SPL of 84dB in a review article (3dB less than claimed by the maufacturer), a more appropriate max. SPL reference value of 101dB for the sensivity measurements would lead to even lower hearing thresholds of 5dB for the 400Hz tone, which is right on the optimistic hearing threshold curve, as shown above...this makes me worry about measuring accuracy, even though the -96dB signal attenuation should be accurate (as this is the threshold when a CD just yields no output any more. That is why I generated the -90 and -96dB test signals before also in 16bits resolution)...

 

[eliash]

Since we hear in Phon (not dB SPL) we might measure 30dB (not dBA) in a quiet room but in the part of the freq, range where we hear actual noise as 'noise' it looks like Ray's measurements show 10dB noise floor isn't that far fetched.
When you overlay Phon curves with the hearing limits Ray's room measures about 10 Phon ?
It basicaly means the room is quiet.

Then it is possible to hear faint tones above that noise floor in the 1-4kHz range thereby concluding that you can hear that low and need a certain dB range.
The trouble is, in practice, music isn't a continuous tone but an addition of many tones spread over a wide band where most of them will always be below 0dBFS so these will be inaudible anyway when peaks (not peaks of individual tones) register at 0dBFS.

To illustrate this I created a file which will show you that 80dB attenuation of music is enough to not hear anything anymore.

Yes, I stumbled about that also yesterday.
Considering real music, e.g. coming from instruments as played in Jazz or Classical, it must be the question, what can be really heared or perceived from them at hearing threshold. It is probably not regular notes played, but some faint noise, which might be interesting for the human perception of the musical presentation. Here I have currently no proof that 24bit are really relevant, even though the ear might be able to go beyond CD resolution...
Having 24bits available, it is of course helpful in recording transients, which would be otherwise clipped rightaway in the studio or in mixing/mastering afterwards, but given a reasonable setup as we all seem to own as friends of audio reproduction, clipping would then occur only later at home in the power amp.

 

[solderdude]

The FLAC file I uploaded (a few posts back in a ZIP file) gives you the answer.
Play the first part as loud as one normally listens (low DR recording) and see what attenuation lets you hear nothing.
That's the actual dynamic range your ears have on your system at your listening position.
Most trust their ears so an easy and quick to do test.

 

[Blumlein 88]

https://www.audiosciencereview.com/...hat-level-is-noise-heard-in-your-system.1013/

I had a thread about this. I put pink noise filtered out below 3 khz and above 5 khz into music files at different levels. The noise was on and off every 2 seconds. You tried different levels of noise to see where you couldn't hear it anymore.

Those files aren't available for download now, but I asked people to set their music at a comfortable listening level and then determine how loud the noise was audible. The answers given here and on another forum clustered around -70 db being where noise wasn't heard. Most people were listening at average levels of 75-80 db on the music. So it is pretty clear you could hear this noise in the most sensitive region somewhere around 10 db SPL. So for that one region we really do need a system quiet enough to reach near the threshold of hearing. Or not too far above it.

 

[edechamps]

Since we hear in Phon (not dB SPL) we might measure 30dB (not dBA) in a quiet room but in the part of the freq, range where we hear actual noise as 'noise' it looks like Ray's measurements show 10dB noise floor isn't that far fetched.

@RayDunzl's measurement does not show a 10 dB noise floor. A standard spectrum display is misleading when it comes to assessing the level of broadband signals (such as noise), because, contrary to pure tones, broadband signals do not land in a single FFT frequency bin. If you change the number of bins, the same amount of energy gets spread over a different number of buckets, and you get different numbers. For example, if @RayDunzl were to change its FFT size from 262144 to, say, 32768, I would expect the entire noise floor shown in the spectrum display to rise by about 9 dB. That's because each FFT bin would then contain 8 times more energy. But of course the physical noise floor itself didn't change just because you tuned a dial on your FFT analyzer!

This means the Y scale on the spectrum display is misleading and doesn't really tell you much in reality, because the numbers depend on the analysis parameters (FFT size). In order to make sense of such numbers they need to be converted into a form that is independent of FFT size; such as by expressing them in dB per Hz (as can be seen on some datasheets).

Then it is possible to hear faint tones above that noise floor in the 1-4kHz range thereby concluding that you can hear that low and need a certain dB range.

Just to make sure this is perfectly clear: you can hear that low and need a certain dB range in that particular frequency band. If your system peak level is 105 dB SPL, and you want to hear a 10 dB SPL tone, that doesn't mean you need 95 dB SPL of dynamic range (which is measured broadband). You can get away with much less than that.

Here I have to correct myself also in the original SPL calcuation. I stated 4dB for the increased hearing distance of 2.5m compared to the speakers SPL spec at 1m distance. It must be 8dB, because the distance ratio is squared (did a multiplication of 10 instread of 20 in the dB calculation).

The inverse square law only applies for freefield conditions, not for reverberant rooms. In a normal room, SPL from a loudspeaker goes down by about 3-4 dB per doubling of the distance, depending on how reverberant your room is. Your original calculation was probably closer to the real number.

 

[Calexico]

I would expect more. I used to do noise an vibration research and an apparently quiet room would be around 30dB. My listening room with just me in it sitting still and all doors and windows shut is 30dBA, 38dB full range, 35 dBC.
In winter the noise of the water running through the heating increases it to 40 dB.
The quietest bits of a Mahler symphony are around 70dB than this when the peaks hit 105dB.
Popular music has a much smaller variation than that

Aren't there harmonics that are lower than 70db?
If you listen a real piano there's is a much more dynamic range and a cd should be able to render it.

 

[Calexico]

Hello i ve got a question. If 16 bit is enough for ear dynamic range. Having more bit can have more precision instead of having more dynamic range? Maybe the ear needs more bits in a normal listening rang ?
Same question for frequency.
If with 44.1khz you can render the whole frequency range. Having higher samplerate can give more precision in the frequency. If it's in the 1khz 2khz range can the frequency be more precise with higher samplerate?
I mean we could get 1.001khz and 1.002 khz more easily for exemple.
Then could it be earable as it's in the very sensitive spot?

 

[edechamps]

Hello i ve got a question. If 16 bit is enough for ear dynamic range. Having more bit can have more precision instead of having more dynamic range? Maybe the ear needs more bits in a normal listening rang ?
Same question for frequency.
If with 44.1khz you can render the whole frequency range. Having higher samplerate can give more precision in the frequency. If it's the 1khz 2khz can the frequency be more precise with higher samplerate?
I mean we could get 1.001khz and 1.002 khz more easily for exemple.
Then could it be earable as it's in the very sensitive spot?

I suspect you are confused as to how digital audio actually works. Please watch this video which should make things clear.

 

[RayDunzl]

For example, if @RayDunzl were to change its FFT size from 262144 to, say, 32768, I would expect the entire noise floor shown in the spectrum display to rise by about 9 dB.

In REW, changing the "FFT Length" parameter does cause a noise level change when viewing the Spectrum display, but when viewing the RTA display, the reported noise level remains the same with different FFT Lengths.

 

[Calexico]

I suspect you are confused as to how digital audio actually works. Please watch this video which should make things clear.

Already watched and it doesn't talk about sensitivity in the range. How much precision of bits and frequency do we need in the range where human earing is very sensitive? That s my question.
I d like an answer like between khz human can detect a difference of 0.001khz or 0.003 khz for ex. and between levels human can detect 0.05 db or 0.9db i don't know.
And then we should see if bit and samplerate are enough.

 

[Krunok]

Aren't there harmonics that are lower than 70db?
If you listen a real piano there's is a much more dynamic range and a cd should be able to render it.

Luckilly for us our ears can easilly recognise different instruments playing the same tone although the harmonics that characterise those instruments are burried deeply in the noise floor.

 

[eliash]

@RayDunzl's measurement does not show a 10 dB noise floor. A standard spectrum display is misleading when it comes to assessing the level of broadband signals (such as noise), because, contrary to pure tones, broadband signals do not land in a single FFT frequency bin. If you change the number of bins, the same amount of energy gets spread over a different number of buckets, and you get different numbers. For example, if @RayDunzl were to change its FFT size from 262144 to, say, 32768, I would expect the entire noise floor shown in the spectrum display to rise by about 9 dB. That's because each FFT bin would then contain 8 times more energy. But of course the physical noise floor itself didn't change just because you tuned a dial on your FFT analyzer!

This means the Y scale on the spectrum display is misleading and doesn't really tell you much in reality, because the numbers depend on the analysis parameters (FFT size). In order to make sense of such numbers they need to be converted into a form that is independent of FFT size; such as by expressing them in dB per Hz (as can be seen on some datasheets).



Just to make sure this is perfectly clear: you can hear that low and need a certain dB range in that particular frequency band. If your system peak level is 105 dB SPL, and you want to hear a 10 dB SPL tone, that doesn't mean you need 95 dB SPL of dynamic range (which is measured broadband). You can get away with much less than that.



The inverse square law only applies for freefield conditions, not for reverberant rooms. In a normal room, SPL from a loudspeaker goes down by about 3-4 dB per doubling of the distance, depending on how reverberant your room is. Your original calculation was probably closer to the real number.

I was just measuring that again with my uncalibrated SPL-meter, since I had doubt about the calculation of the of the max. SPL level myself. Actually it is very difficult to obtain a stable signal here at the listening postion with the pure 400Hz tone, so I changed to a Critical Band Noise #5 signal (450Hz center, 110Hz BW, which I created some time ago). That yielded stable readings (e.g. when myself moving through the room), though switching on and off the left or right speaker don´t show the same levels individually, nor the field summation at the listening point is 6dB. Nevertheless there is a reasonably stable level existent, so I continued testing out. That CBNoise signal was designed to have -20dB average level, as Foobar shows on the VU-meter. Setting the amp gain like before (testing the 400Hz sine wave tone) I could reduce the foobar gain (which I can do it in 1dB steps this with a Pronto IR-remote controlling the netbook via Girder).
Switching on and off the signal at -78db yielded the least faintly hearable output. Together with the -20dB CBN-signal this makes 98dB below an newly assumed 105dB max. SPL, which happens to be exactly in the range of the pure 400Hz sine wave measured before, namely 7dB SPL...

 

[Krunok]

I d like an answer like between khz human can detect a difference of 0.001khz or 0.003 khz for ex. and between levels human can detect 0.05 db or 0.9db i don't know.
And then we should see if bit and samplerate are enough.

Answer to those questions were obtain by blind hearing tests, not by physics/medicine. The same tests confirmed that CD quality is fine and that higher sample rates don't really bring any advantage. You can raise bit depth to 24 to avoid collecting noise during the recording/mastering process, but that's about it. 16/44.1 format has not been chosen "out of the blue".

 

[Blumlein 88]

Already watched and it doesn't talk about sensitivity in the range. How much precision of bits and frequency do we need in the range where human earing is very sensitive? That s my question.
I d like an answer like between khz human can detect a difference of 0.001khz or 0.003 khz for ex. and between levels human can detect 0.05 db or 0.9db i don't know.
And then we should see if bit and samplerate are enough.

For humans normally a .3% difference in two frequencies will be heard. So at 3000 hz that would be 9 hz like 3000 hz and 3009 hz. Now beating between two tones at the same time can be heard closer together, but that is a different thing than you are asking. And CD can handle that with more precision than humans.

We know that .2 db level difference will corrupt blind testing. .1 db will not. So that is somewhere near the limit.

16 bit 44.1 khz can easily cover either one of these with a huge margin.

EDIT: I originally typed in 3%, but the proper number is .3% which is 9 hz for a 3000 hz tone.

 

[edechamps]

In REW, changing the "FFT Length" parameter does cause a noise level change when viewing the Spectrum display, but when viewing the RTA display, the reported noise level remains the same with different FFT Lengths.

Yes, I believe that's because the RTA display converts the values to total amount of energy per octave (or some fraction of an octave), thus solving the problem as long as the reader is aware of what's happening and can convert the numbers to whatever basis is appropriate for the point being made.

Already watched and it doesn't talk about sensitivity in the range. How much precision of bits and frequency do we need in the range where human earing is very sensitive? That s my question.

That would be a valid question for an undithered signal. All audio signals are dithered. The point of dither is precisely to take such imprecisions, which can result in offensive distortion (quantization distortion), and turn them into broadband noise instead (which is benign in comparison). In a dithered signal, these imprecisions you're referring to are the dither noise itself. If you ignore the noise, you end up with infinite precision.

I was just measuring that again with my uncalibrated SPL-meter, since I had doubt about the calculation of the of the max. SPL level myself. Actually it is very difficult to obtain a stable signal here at the listening postion with the pure 400Hz tone, so I changed to a Critical Band Noise #5 signal (450Hz center, 110Hz BW, which I created some time ago).

If you want to calibrate to a reference SPL, it's easier to use a broadband signal such as white noise. It will average away issues affecting specific frequencies (resonances, reflections, etc.), providing a more stable reading. I would also band-limit the noise around 1 kHz or so, to centre on the frequency range where the SPL meter is most accurate.

 

[Calexico]

For humans normally a 3% difference in two frequencies will be heard. So at 3000 hz that would be 9 hz like 3000 hz and 3009 hz. Now beating between two tones at the same time can be heard closer together, but that is a different thing than you are asking. And CD can handle that with more precision than humans.

We know that .2 db level difference will corrupt blind testing. .1 db will not. So that is somewhere near the limit.

16 bit 44.1 khz can easily cover either one of these with a huge margin.

Thank you

 

[andreasmaaan]

Here I have to correct myself also in the original SPL calcuation. I stated 4dB for the increased hearing distance of 2.5m compared to the speakers SPL spec at 1m distance. It must be 8dB, because the distance ratio is squared (did a multiplication of 10 instread of 20 in the dB calculation). Together with a really measured speaker SPL of 84dB in a review article (3dB less than claimed by the maufacturer), a more appropriate max. SPL reference value of 101dB for the sensivity measurements would lead to even lower hearing thresholds of 5dB for the 400Hz tone, which is right on the optimistic hearing threshold curve, as shown above...this makes me worry about measuring accuracy, even though the -96dB signal attenuation should be accurate (as this is the threshold when a CD just yields no output any more. That is why I generated the -90 and -96dB test signals before also in 16bits resolution)...

Your original method was incorrect, but in fact I think it led to a more correct result than your current method, because you also overlooked the effect of reflections.

In a typical domestic room, the "critical distance" tends to be around 75-150cm from the speaker. This is the distance beyond which the SPL of reflected sounds is greater than the SPL of the direct sound.

In practical terms, this means that if a speaker produces 84dB at 1W/1m, in a real room it is likely to produce about the same SPL at any distance beyond 1m, too (of course this will be frequency-dependent*). So your original estimate of 108dB at the listening position for an 0dBfs input signal is likely to be closer to the reality than your revised figure of 101dB IMHO.

*For this reason, I suggest you do the same experiment with noise rather than a single 400Hz tone.