Quantcast
My null test 88.2Kh 24 bit and 44.1Kh 16 bit - Gearspace.com
The No.1 Website for Pro Audio
My null test 88.2Kh 24 bit and 44.1Kh 16 bit
Old 17th August 2018
  #1
Gear Addict
 
🎧 5 years
My null test 88.2Kh 24 bit and 44.1Kh 16 bit

I really don't want to reawaken the tired old arguments about high res vs low res audio. But to satisfy my own curiosity and because I was feeling a bit geeky today, I conducted a little null test experiment.

On my own system I can't do a true null test between 88.2/24 bit and 44.1/16 bit but I did a null test as close as I could get.

I took an original 88.2/24 bit recording and loaded it into Izotope Rx. One can see the high frequencies displayed above 20k. Not sure how that is possible because I used ribbons but apparently there is still something up there.





In this order, I down-sampled to 44.1, reversed the phase, dithered to 16 bit, up-sampled back to 88.2Khz and truncated from 32 bit float to 24 bit fixed.




I loaded into Samplitude the original 88.2/24 bit version with the new phased reversed one that went through two re-samplings, turned everything way up and listened on head phones- nothing, nada, but some white noise. At normal listening levels I couldn't hear any white noise.




I exported the mix from both tracks and loaded it back into Izotope Rx to see what I could see. Obviously there is still some high frequency content and the dither noise is observable. I imagine there is some quantization distortion from the truncation of 32 bit float to 24 bit fixed and perhaps some ringing from the sample rate conversion. Who knows.




One can see more information in the left channel, most likely because the drums were situated off to the left in the recording so the cymbals created some high frequency info.




I imagine there is a better way to do this but am I wrong to say that, in the audible range (20hz-20Khz) a down-sampled 44.1/16 bit and an original 88.2/24 bit version are, for all intents and purposes, identical. Even after two sample rate conversions and dithering to 16 bit and back up, they null except some high frequency info and noise.

Maybe this is all a no brainer, but sometimes I have to do these things for myself! haha

Cheers
Old 17th August 2018
  #2
Lives for gear
 
Verified Member
🎧 10 years
Did you blind A/B test the two files ?
Old 17th August 2018 | Show parent
  #3
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by Saxnscratch ➑️
Did you blind A/B test the two files ?
I did not blind test them - guess I could try that. Are you thinking that the noise that didn't null would make a difference?
Old 17th August 2018 | Show parent
  #4
Lives for gear
 
Verified Member
🎧 10 years
Quote:
Originally Posted by shosty ➑️
I did not blind test them - guess I could try that. Are you thinking that the noise that didn't null would make a difference?
I don't know, tell me
And Try 96 -> 44.1 ->96 (with modern src, it shouldn't really make a difference)

I'm curious of your observations
Old 17th August 2018 | Show parent
  #5
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by Saxnscratch ➑️
I don't know, tell me
And Try 96 -> 44.1 ->96 (with modern src, it shouldn't really make a difference)

I'm curious of your observations
Ok, sure...I'll give it a try this weekend. I'll see if I can get my wife to help.
Old 18th August 2018
  #6
Lives for gear
 
scraggs's Avatar
 
4 Reviews written
🎧 15 years
Quote:
Originally Posted by shosty ➑️
I imagine there is a better way to do this but am I wrong to say that, in the audible range (20hz-20Khz) a down-sampled 44.1/16 bit and an original 88.2/24 bit version are, for all intents and purposes, identical.
IMO, no you are not wrong. i did a similar test to yours recently....took an 88.2/32 file, downsampled to 44.1, saved that as a 32fp file, upsampled back to 88.2, saved that again as a 32fp file.

flipped phase on the down/up file, played it against the original, and my findings were the same as yours, no difference except for some very low level high freqs above 22k.

so i suppose a dog in an extremely quiet room might be able to hear a difference, but humans, not so much.
Old 18th August 2018 | Show parent
  #7
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by scraggs ➑️
IMO, no you are not wrong. i did a similar test to yours recently....took an 88.2/32 file, downsampled to 44.1, saved that as a 32fp file, upsampled back to 88.2, saved that again as a 32fp file.

flipped phase on the down/up file, played it against the original, and my findings were the same as yours, no difference except for some very low level high freqs above 22k.

so i suppose a dog in an extremely quiet room might be able to hear a difference, but humans, not so much.
Interesting. What I figure is, if any information was lost in the down sample, it cannot be regained in the up sample (hence the high frequencies don't null). So it seems that the difference in sample rates between 44.1 and 88.2 have no bearing on any other part of the audio outside of the high frequencies (assuming a good SRC).
Old 18th August 2018
  #8
Lives for gear
 
scraggs's Avatar
 
4 Reviews written
🎧 15 years
right. you can't regain anything by upsampling, it's just making it easy to do a null test.
Old 18th August 2018 | Show parent
  #9
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by scraggs ➑️
right. you can't regain anything by upsampling, it's just making it easy to do a null test.
Ok, not being the professional here, just a lowly cellist, I just want to be sure I am on track! haha
Old 18th August 2018
  #10
Lives for gear
 
scraggs's Avatar
 
4 Reviews written
🎧 15 years
you're doing great!

cellists lowly? i'm a drummer.....
Old 18th August 2018 | Show parent
  #11
Gear Nut
 
moostapha's Avatar
 
🎧 5 years
Quote:
Originally Posted by shosty ➑️
Ok, sure...I'll give it a try this weekend. I'll see if I can get my wife to help.
You don't necessarily need help. ABXTester exists on OS X. I'm sure either there's a windows version or a similar piece of software for Win.

Also, yeah...numerous studies have shown the same thing as your findings. There are still reasons (perhaps purely theoretical) to work at higher sample rates, such as basically getting ~2x-4x oversampling from your plugins even if the plugins don't support it. Of course, the trade-off is mostly file size and the need for computing power, and a few claims I've heard but not seen proven. But, none of those are the end of the world.
Old 18th August 2018
  #12
Gear Head
 
🎧 5 years
I don’t understand what this is accomplishing. The whole beef with sample rate is how a converter is able to receive an input from an analog source and output that signal in the digital format to a computer. And not just in the excesses of the sample rate, but in the audible range, between 20 hz and 20 khz. Of course a signal that is already 1’s and 0’s at 88.2, downsampled and then upsamples again will show nothing in a null test. The converter itself isn’t actually doing anything. Just a bunch of math equations in the digital realm. ( which a computer is really good at doing)
Old 18th August 2018 | Show parent
  #13
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by scraggs ➑️
you're doing great!

cellists lowly? i'm a drummer.....
Ah, but you are an audio pro (I took a peek at your website). I used to live in Boston, BTW.

Thank you!
Old 18th August 2018 | Show parent
  #14
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by moostapha ➑️
You don't necessarily need help. ABXTester exists on OS X. I'm sure either there's a windows version or a similar piece of software for Win.

Also, yeah...numerous studies have shown the same thing as your findings. There are still reasons (perhaps purely theoretical) to work at higher sample rates, such as basically getting ~2x-4x oversampling from your plugins even if the plugins don't support it. Of course, the trade-off is mostly file size and the need for computing power, and a few claims I've heard but not seen proven. But, none of those are the end of the world.
Ok, cool. I'll check it out. Thanks.

For serious projects I usually record at higher sample rates for the reasons you mention. But it's good to know that the conversion process does not degrade the audio.
Old 18th August 2018 | Show parent
  #15
Gear Nut
 
moostapha's Avatar
 
🎧 5 years
Quote:
Originally Posted by shosty ➑️
Ok, cool. I'll check it out. Thanks.

For serious projects I usually record at higher sample rates for the reasons you mention. But it's good to know that the conversion process does not degrade the audio.
Well, in a theoretical sense, assuming that your mics and everything else can record past 22k, it does. Itswjust very different from, say, the mp3 debate...which is apparently still a point of contention for some people. Because there is strong scientific evidence that humans can't hear the degradation.
Old 18th August 2018 | Show parent
  #16
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by kretzmonster ➑️
I don’t understand what this is accomplishing. The whole beef with sample rate is how a converter is able to receive an input from an analog source and output that signal in the digital format to a computer. And not just in the excesses of the sample rate, but in the audible range, between 20 hz and 20 khz. Of course a signal that is already 1’s and 0’s at 88.2, downsampled and then upsamples again will show nothing in a null test. The converter itself isn’t actually doing anything. Just a bunch of math equations in the digital realm. ( which a computer is really good at doing)
This simple test seems to me, anyway, to show that no audio is lost in the down sample in the audible range. Kind of good info to have for a variety of reasons, I think.

Should I record at 44.1, since it is the final format anyway, or a higher sample rate? Well, at least I know the conversion won't hurt anything - just a loss of high frequency. And it's pretty cool to know sample rate conversion is at a level today that is quite stellar. If done with a good SRC, a 44.1 recording should be virtually indistinguishable from the original higher sample rate.

BTW, are you saying that hardware cannot do as good a job at SRC as software? I'm not sure I am really understanding your post. So if I recorded the same group and outputted at 44.1 and 88.2 simultaneously from the same converter, I wouldn't get the same null by up sampling the 44.1 or down sampling the 88.2? I suppose that would be an interesting test - I'm sure it's been done.

Last edited by shosty; 18th August 2018 at 05:15 AM.. Reason: Asked a question
Old 18th August 2018 | Show parent
  #17
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by moostapha ➑️
Well, in a theoretical sense, assuming that your mics and everything else can record past 22k, it does. Itswjust very different from, say, the mp3 debate...which is apparently still a point of contention for some people. Because there is strong scientific evidence that humans can't hear the degradation.
Yes, agree. The conversion just doesn't seem to degrade the audio up to 22k, I should say. And it doesn't really degrade the audio above 22k, does it? It just leaves it out. Which is different, as you say, from the mp3 debates, which leaves out content within the audible range.

We agree, I think...

Last edited by shosty; 18th August 2018 at 05:03 AM..
Old 18th August 2018 | Show parent
  #18
Gear Nut
 
moostapha's Avatar
 
🎧 5 years
Quote:
Originally Posted by shosty ➑️
Yes, agree. The conversion just doesn't seem to degrade the audio up to 22k, I should say. And it doesn't really degrade the audio above 22k, does it? It just leaves it out. Which is different, as you say, from the mp3 debates, which leaves out content within the audible range.

We agree, I think...
Yes.

I realized what I typed was ambiguous. Some tracks/signals make it easier/harder than others, but mp3, aac, etc. is detectable at least a significant amount of the time. SRC or "HD" sample rates...the last time I went to an audiologist, he was amazed at my hearing for my age, and those results show that there's no way I could hear it. And I'm not convinced anyone genetically normal could.

Now...distortion from a converter that cuts corners at higher sample rates...that happens and is audible. But, it's also been a while since that was a widespread problem.
Old 18th August 2018 | Show parent
  #19
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by moostapha ➑️
Yes.

I realized what I typed was ambiguous. Some tracks/signals make it easier/harder than others, but mp3, aac, etc. is detectable at least a significant amount of the time. SRC or "HD" sample rates...the last time I went to an audiologist, he was amazed at my hearing for my age, and those results show that there's no way I could hear it. And I'm not convinced anyone genetically normal could.

Now...distortion from a converter that cuts corners at higher sample rates...that happens and is audible. But, it's also been a while since that was a widespread problem.
No problem. The last I checked I could hear up to 15Khz, VERY faintly. But that was a couple years ago. I'm not so sure anymore.

Part of why I did the test was because I wanted to see how good the sample rate conversion is today. I remember back in the 90s when, to get a nice smooth conversion without audible edge, you had to go to the mastering studio. I suppose that is bad news on this mastering sub forum! But anyway, I feel pretty rest assured that sample rate conversion is quite stellar today.
Old 18th August 2018 | Show parent
  #20
Gear Nut
 
moostapha's Avatar
 
🎧 5 years
Quote:
Originally Posted by shosty ➑️
No problem. The last I checked I could hear up to 15Khz, VERY faintly. But that was a couple years ago. I'm not so sure anymore.

Part of why I did the test was because I wanted to see how good the sample rate conversion is today. I remember back in the 90s when, to get a nice smooth conversion without audible edge, you had to go to the mastering studio. I suppose that is bad news on this mastering sub forum! But anyway, I feel pretty rest assured that sample rate conversion is quite stellar today.
SRC Comparisons

RX stacks up pretty well, FWIW. It's gotten cheaper to do math in the last 20 years.
Old 18th August 2018
  #21
Tokyo Dawn Labs
 
FabienTDR's Avatar
 
Verified Member
🎧 5 years
While the result of optimal src is no more than a low pass filter plus noise, testing these is slightly more elaborate.

Distortion and aliasing won't really be obvious in your test setup, sadly that's exactly where issues appear.

Sine sweeps covering the whole source bandwidth would yield more sensible results, also audibly, without any AB.
Old 18th August 2018 | Show parent
  #22
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by moostapha ➑️
SRC Comparisons

RX stacks up pretty well, FWIW. It's gotten cheaper to do math in the last 20 years.
Yes, it does really well! Very cool set of tests. Thanks.
Old 18th August 2018 | Show parent
  #23
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by FabienTDR ➑️
While the result of optimal src is no more than a low pass filter plus noise, testing these is slightly more elaborate.

Distortion and aliasing won't really be obvious in your test setup, sadly that's exactly where issues appear.

Sine sweeps covering the whole source bandwidth would yield more sensible results, also audibly, without any AB.
Yes, I agree that the tests done by Infinite Wave at src.infinitewave.ca are going to reveal more. I suppose if there was any aliasing in my test, it was masked by the white noise and I couldn't hear it. On a slightly different point, if you look at their results, virtually all show artifacts well below audibility. Isn't something like -80db below signal is considered inaudible?
Old 18th August 2018
  #24
Lives for gear
 
Trakworx's Avatar
 
Verified Member
3 Reviews written
🎧 10 years
shosty - Years ago I did similar tests with similar results. It is interesting and good to know that the audible range information in the files themselves is preserved when downsampling to 16-44.

That leaves the DAC as the remaining variable when considering whether playing back audio at different SRs actually sounds different. Much harder to test that, and results would only apply to the model of DAC tested...
Old 18th August 2018 | Show parent
  #25
Lives for gear
 
scraggs's Avatar
 
4 Reviews written
🎧 15 years
Quote:
Originally Posted by FabienTDR ➑️
Distortion and aliasing won't really be obvious in your test setup, sadly that's exactly where issues appear.
Fabien, could you elaborate on this please? Wouldn't distortion/aliasing show up in a null test?
Old 18th August 2018
  #26
Tokyo Dawn Labs
 
FabienTDR's Avatar
 
Verified Member
🎧 5 years
Arbitrary music signals do not fully stimulate/provoke worst case situations. Thats the main problem. This is particularly true with old orchestral recordings. A fat lofi hiphop or dubstep track could sound fresher and deeper with subtle aliasing and quantization artefacts. Music is a serious bottleneck "in the lab"

Music also tends to summon subjective memories that seriously distort perception.

Null tests in this specific situation have an issue with timing. Some src filter and configuration might well shift the signal by a half sample further than the other test case. This in turn could easilly provoke false positives. However, as long you can exclude timing issues and use a complete stimulus e.g. sweeps or dirac, it' s totally ok to watch the difference signal.

Imho logarithmic sweeps with alias free waveforms are much more informative and audible. Dirac pulses spaced about 2 sec are great representatives for the fastest, highest quality percussive signals and how they behave through the system. These two approaches deliver the full picture. Both audibly and measurably.
Old 18th August 2018 | Show parent
  #27
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by Trakworx ➑️
shosty - Years ago I did similar tests with similar results. It is interesting and good to know that the audible range information in the files themselves is preserved when downsampling to 16-44.

That leaves the DAC as the remaining variable when considering whether playing back audio at different SRs actually sounds different. Much harder to test that, and results would only apply to the model of DAC tested...
Yes, exactly. Very good distinction. Thanks.
Old 18th August 2018
  #28
Gear Addict
 
🎧 5 years
The difference will be only with high sample rates vs low sample rates while recording, mixing, processing, mastering... what I see in this picture is mbit+ (?) dithering and frequency content above circa 22kHz, which should be shown in null test after "ideal" src conversion with sharpest high cut filter. If it is almost ideal, you won't see aliasing.
Old 19th August 2018 | Show parent
  #29
Gear Addict
 
🎧 5 years
Quote:
Originally Posted by Stereo Flux ➑️
The difference will be only with high sample rates vs low sample rates while recording, mixing, processing, mastering... what I see in this picture is mbit+ (?) dithering and frequency content above circa 22kHz, which should be shown in null test after "ideal" src conversion with sharpest high cut filter. If it is almost ideal, you won't see aliasing.
I believe I used very close to the ideal filter. And the dithering is one of the standard presets for mbit in Rx.
πŸ“ Reply

Similar Threads

Thread / Thread Starter Replies / Views Last Post
replies: 217 views: 51963
Avatar for Leverson
Leverson 7th June 2014
replies: 97 views: 13366
Avatar for tboston007
tboston007 29th September 2014
Topic:
Post Reply

Welcome to the Gearspace 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…

Forum Jump
Forum Jump