Audibility thresholds of S/N - signal to noise - test

[pma]

SOTA electronic devices like DACs, preamplifiers and amplifiers have now incredibly high S/N ratio. The limiting factors are in fact S/N of recordings and ambient noise in a listening room. Would you like to check how much S/N you can reliably hear in your listening conditions? Below is a link to test files. They contain a sine 200Hz signal and the same 200Hz sine with added white noise, with rms level of -20dB, -30dB, -40dB, -50dB, -60dB and -70dB relative to 200Hz sine rms level.

https://pmacura.cz/testsignaltonoise.zip

In case you claim to hear -70dB S/N, please add your ABX record.

 

[JSmith]

The noise is quite audible from -20dB to -50dB, it's slightly audible @-60dB, but @-70dB it's not easily discernible.

That said, I will need to try this on my full home system and also HP's.

Regardless of audibility... no one wants noise though, audible or not, so unsure of the main point here apart from when noise is too high it's audible.


JSmith

 

[pma]

Regardless of audibility... no one wants noise though, audible or not, so unsure of the main point here apart from when noise is too high it's audible.

That's right, but do you need S/N 140dB if you already have S/N 110dB? The question is when the race looses practical sense. And, very small amount of noise may act as a dither and in fact lead to subjectively "better" sound. So, is technically better always better for the purpose?

 

[olieb]

Maybe you should have added a -80dB option. Not because it is audible but to give the golden ears a chance to out themselves.
Would be interesting to do the same test with a 100Hz tone, too.

That's right, but do you need S/N 140dB if you already have S/N 110dB?

For me for all practical reasons s/n of 90...100dB (16bit) is certainly enough. But as I use digital volume control some more for headroom is nice.
EDIT: My limit is -60dB (just so) like all others so far.

 

[DanielT]

Thanks for putting this test together PMA.

The noise is quite audible from -20dB to -50dB, it's slightly audible @-60dB, but @-70dB it's not easily discernible.

That said, I will need to try this on my full home system and also HP's.

Regardless of audibility... no one wants noise though, audible or not, so unsure of the main point here apart from when noise is too high it's audible.


JSmith

Same for me, but it wasn't so easy to detect at -60dB. I could barely detect it through the speakers, faint, but for some reason not through my headphones.

 

[Drengur]

60 in my not so top-of-the-line office headphones.

 

[Sokel]

-50 at first try, -60 at the second, all through speakers.

 

[MaxwellsEq]

Results so far seem to confirm we're more sensitive to noise than we are to distortion.

 

[amirm]

This is not how you test for maximum audibility of channel noise. My reel to reel is pretty quiet while playing music. But gaps between tracks have clear audible noise. In this day and age, we want total silence when nothing is playing.

 

[olieb]

Results so far seem to confirm we're more sensitive to noise than we are to distortion.

Interesting idea, but my guess is that it depends on the kind of distortion. H2 and H3 are different than H7 or H13.
The noise in the files is audible mainly in the 2-4kHz range from max hearing sensitivity in this range.
If I add H9 to 200 Hz (resulting in distortion at 1.8kHz) my audibility threshold is about 0.2%≈-55dB. With H17 it will probably be close to -60dB, but I cannot check that with REW.
So the difference in masking of distortion and noise might not be so big after all.

This is not how you test for maximum audibility of channel noise.

Sure, this is not a test for noise audibility but for masking. For low volume passages (or even silence) one wants more s/n of the system.

 

[AnalogSteph]

I could barely detect it through the speakers, faint, but for some reason not through my headphones.

That's because our auditory processing is trained to deal with localized noise sources with some degree of correlation between both ears. Any uncorrelated random noise just adds to the internally generated racket from blood circulation and whatnot, which tends to get filtered out. As a result, you can generally pick up white noise on speakers down to about +4 dB SPL, while on headphones it'll tend to disappear somewhere around +20. Should be a bit higher still on IEMs.

You can estimate required dynamic range as follows:
Required S/N at normal volume
+20-30 dB of volume headroom
+accounting for level mismatch (e.g. DAC max output > power amp clipping level, speaker sensitivity etc.)

In real life that'll net you 95-110 dB instantaneous plus level mismatch, but even that won't generally add more than 6-12 dB (unless you need to accommodate both sensitive monitors and Hollywood studio levels, in which case it would be more like 24 dB). So odds are that anything from the low 120s up in a DAC is more than adequate, and if you're not a loud listener you may get away with something in the low 100s.

 

[MCH]

This is not how you test for maximum audibility of channel noise. My reel to reel is pretty quiet while playing music. But gaps between tracks have clear audible noise. In this day and age, we want total silence when nothing is playing.

This is exactly my problem with noise. Cannot stand it when nothing is playing. While sound is playing, sure, don't need ultra low levels.

 

[Mike123]

When I tested the sensitivity of my ears to quiet sounds, I modulated the sound. First, the frequency is 3000 Hz. Second, I stopped/started the generator with the mouse. The ears hear changes in sound better than continuous sound. My limits are -85 dBFS from the listening point and -104 dB near the speaker. I do not think that I need -140 dB SNR. But I do not want to have equipment higher than -110 dB.

Yes, according to the test, in a regular USB headset at work I hear -60 dB. I don't hear the difference between -70dB and pure. But on both I hear playback noise, which is not present in the pause. Using 44100/16 is not the best idea .

 

[Mike123]

I read an interview with a sound engineer who releases vinyl a long time ago. There was information there that in regular vinyl recording they tried not to use a dynamic range of more than 40 dB. Precisely because quiet sounds are completely masked by loud ones and there is no point in preserving them. I checked several of my vinyl digitalizations. The noise between songs is -50..-54 dB. They do not interfere with listening to music at all.

 

[DanielT]

That's because our auditory processing is trained to deal with localized noise sources with some degree of correlation between both ears. Any uncorrelated random noise just adds to the internally generated racket from blood circulation and whatnot, which tends to get filtered out. As a result, you can generally pick up white noise on speakers down to about +4 dB SPL, while on headphones it'll tend to disappear somewhere around +20. Should be a bit higher still on IEMs.

You can estimate required dynamic range as follows:
Required S/N at normal volume
+20-30 dB of volume headroom
+accounting for level mismatch (e.g. DAC max output > power amp clipping level, speaker sensitivity etc.)

In real life that'll net you 95-110 dB instantaneous plus level mismatch, but even that won't generally add more than 6-12 dB (unless you need to accommodate both sensitive monitors and Hollywood studio levels, in which case it would be more like 24 dB). So odds are that anything from the low 120s up in a DAC is more than adequate, and if you're not a loud listener you may get away with something in the low 100s.

Thanks for the reply. That was interesting.

When I tested the sensitivity of my ears to quiet sounds, I modulated the sound. First, the frequency is 3000 Hz. Second, I stopped/started the generator with the mouse. The ears hear changes in sound better than continuous sound. My limits are -85 dBFS from the listening point and -104 dB near the speaker. I do not think that I need -140 dB SNR. But I do not want to have equipment higher than -110 dB.

Yes, according to the test, in a regular USB headset at work I hear -60 dB. I don't hear the difference between -70dB and pure. But on both I hear playback noise, which is not present in the pause. Using 44100/16 is not the best idea .

My limits are -85 dBFS from the listening point and -104 dB near the speaker. Hell, thats loud

I measured via a free dB app. When I tested: Around 49 dB, about 2.5 m / 8ft from the speakers. With music it was 62 dB without changing the volume then.
I don't know if that app is that accurate, I must add.

Plus I don't know anything about the signal strengths into DAC PMA files vs Spotify. For convenience done via Bluetooth (LADC)

 

[Mike123]

Hell, thats loud

If you don't confuse SPL and DBFS.

 

[DanielT]

If you don't confuse SPL and DBFS.

You're right. The not insignificant detail that I missed. Sorry.

 

[pma]

Thanks to all who have tried the test.

 

[pma]

But gaps between tracks have clear audible noise.

Noise is always important only when related to useful content. That's why I keep the same 200Hz tone in all samples. Otherwise one would be tempted to turn up volume until he would hear the noise. It does not make sense.

One can also turn volume down and repeat the same test with posted files. This says how S/N audibility is related to programme volume.
Regarding 200Hz, I have chosen it because the ear is less sensitive than to 1kHz - 3kHz range of frequencies and thus the noise is more audible. Of course the speaker is to be able to reproduce 200Hz, cell phones are prohibited in this test, it would be cheating.

 

[pma]

I have just banned the poll option of multiple S/N choices. I have observed that one voter checked all options and it has made the poll confusing.