An Argument For Recording At High Res...

[insub]

Originally Posted by clepsydrae

Sorry, could you clarify a little what you mean here?

Basically, I created a SINE sweep from 20 Hz - 10 kHz and automated the output gain of the tone generator to roughly output -3 dB/oct to mimic Pink Noise.

I sent a separate tone to ReaComp's sidechain input to maintain a -6 dB reduction with a 4:1 ratio (simulate mixing condition) because the GR would not remain static with the source signal. I set all other ReaComp settings to ZERO, including the RMS window.

At 0 ms RMS, the noise is everywhere in the spectrum and very clearly audible. But, even at 3-4 ms the noise is all but gone. My conclusion is that the default RMS window of 5 ms is not just to simulate an analog compressor (thought I read somewhere fastest hardware compressor circuits were approx. 5 ms reaction window) but rather to combat the extreme aliasing that occurs below 5 ms.

That doesn't mean there is no aliasing occurring. However the aliasing that remains is not solved with higher samplerate because it is folding back from the subharmonics extending below 0 Hz. These start folding back at frequencies at & below 1 kHz fundamental and were about -45 dB for the first subharmonic which eventually extends above the fundamental until it mates up with the first harmonic but doesn't appear to add together in amplitude with the first harmonic.

Additionally, I had never considered that subharmonics were generated from compression. In my test, ReaComp appears to generate equal subharmonics to the generated harmonics. Three harmonics in either direction, above & below the fundamental source tone. Each very low in amplitude. About -45 dB for the first, -70 dB for the 2nd, and I did not measure the level of the 3rd.

The GIF is very telling. I hope I can share it with you all soon. I tried PhotoBucket, but they do not allow GIFs to be uploaded.

[insub]

Originally Posted by karbomusic

Nothing really other than the extreme convenience of never clipping while in the DAW for the most part. I have all my render/glue settings etc. set to 32f which means no matter how badly I dork up, I can recover it after the fact - until it's time for the final mix render that gets released.

And, do you find that the higher bit depth files do not have the same negative impact in processing and storage space as 96 kHz samplerate?

I truly do not know what difference the bit depth makes on the file size. I assume higher bit depth = larger file, but not to what degree.

[karbomusic]

Originally Posted by insub

And, do you find that the higher bit depth files do not have the same negative impact in processing and storage space as 96 kHz samplerate?

I truly do not know what difference the bit depth makes on the file size. I assume higher bit depth = larger file, but not to what degree.

I don't remember the file size difference (I think it's bigger) but it only applies to my master render and files I apply/render/glue (original recordings stay at 24). It's for one reason and one reason only, I can clip the utter shit out of something by mistake, render that via some process such as glue, then recover it by just reducing the volume because the clipping point for 32f is something larger than +1000dB.

We get that safety by default when mixing in 32 bit float but this just extends that safety to any rendered files.

[insub]

Originally Posted by insub

Basically, I created a SINE sweep from 20 Hz - 10 kHz and automated the output gain of the tone generator to roughly output -3 dB/oct to mimic Pink Noise.

I sent a separate tone to ReaComp's sidechain input to maintain a -6 dB reduction with a 4:1 ratio (simulate mixing condition) because the GR would not remain static with the source signal. I set all other ReaComp settings to ZERO, including the RMS window.

At 0 ms RMS, the noise is everywhere in the spectrum and very clearly audible. But, even at 3-4 ms the noise is all but gone. My conclusion is that the default RMS window of 5 ms is not just to simulate an analog compressor (thought I read somewhere fastest hardware compressor circuits were approx. 5 ms reaction window) but rather to combat the extreme aliasing that occurs below 5 ms.

That doesn't mean there is no aliasing occurring. However the aliasing that remains is not solved with higher samplerate because it is folding back from the subharmonics extending below 0 Hz. These start folding back at frequencies at & below 1 kHz fundamental and were about -45 dB for the first subharmonic which eventually extends above the fundamental until it mates up with the first harmonic but doesn't appear to add together in amplitude with the first harmonic.

Additionally, I had never considered that subharmonics were generated from compression. In my test, ReaComp appears to generate equal subharmonics to the generated harmonics. Three harmonics in either direction, above & below the fundamental source tone. Each very low in amplitude. About -45 dB for the first, -70 dB for the 2nd, and I did not measure the level of the 3rd.

The GIF is very telling. I hope I can share it with you all soon. I tried PhotoBucket, but they do not allow GIFs to be uploaded.

I still cannot get the GIF to upload. So, here is the project file.
https://stash.reaper.fm/v/33516/Aliasing%20ReaComp.rpp

Unfortunately, you need Dead Duck's DD SigGen (FREE) for it to work because I did not know how to make a smooth, controllable (-3dB/oct) sweep with REAPER features. JS: Tone Generator changes musical pitches in steps, as far as I can tell. Get Dead Duck free FX pack here: http://deadducksoftware.blogspot.rs/

[clepsydrae]

Originally Posted by cyrano

There are hardly any audio interfaces that are capable of producing a 32 bit stream. There is one that I know of, but I've never seen it used by anyone. It costs between 6 and 8.000 euro's IIRC.

Yeah, i think that's the one I was alluding to in my post. And I assume it's 32bit int?

So I'm figuring I must be misunderstanding something here?

I think the tester wanted to test high sample rates but muddied the water with the 32bit-float bit depth thing.

That's why I don't see much use in this test. And it shows serious questions about the background knowledge of these testers. If there's one thing to avoid, it is resampling. If SRC occurs, you'll probably never know what you're testing.

Yeah. I think the useful test would be two files, one 44.1/24 and the other 192k/24. Set your interface to 192k and play them back. The 44.1 will get upsampled (and if you care, you can pre-upsample it with sox or whatever). That would seem a sensible comparison, no? And the one we're hoping some helpful person posts, demonstrating the audible difference that is claimed by some.

I should point out something else I noticed: in a reaper project with two clips of dissimilar sample rates (the two demo files from the test in question), I tried nulling them. Not scientific, I know, but I did it. The interesting part was that as the playback looped, the resulting output changed dramatically with each repeat of the loop. It was as if the resampling was going in and out of phase or something. Sometimes there was little-to-no output, sometimes it was pretty loud (considering it should have been ~silent). When I resampled with sox and nulled, I got the results I described of ~silence.

...that's just a caution in case anyone thinks they can throw dissimilar sample rate files into reaper and get reliable/repeatable on-the-fly resampling during null testing.

[cyrano]

Originally Posted by insub

However, there are still enough other minuscule reasons for me to still want to mix at 96 kHz. They are primarily technical reasons, not necessarily audible reasons.

Have you considered recording at 48 kHz and upsampling?

I have a sneaky suspicion this might even be better. I've heard the argument that 96 is better for some VST's and I think there's some sense in it. Never tried to test it, though, as it doesn't keep me awake at night.

But I also use my FF400 at 192 for measurements. Everything I've had before wasn't really up to that. The reason being that some interfaces don't even pass audio over 20 kHz, even if they announce 96 kHz sample rate. They seem to have fixed analog filters. And some others aren't very clean in what they pass at very high frequencies, but that depends on the environment. And you don't want noise you can't hear, do you?

[insub]

Originally Posted by cyrano

Have you considered recording at 48 kHz and upsampling?

I have a sneaky suspicion this might even be better. I've heard the argument that 96 is better for some VST's and I think there's some sense in it. Never tried to test it, though, as it doesn't keep me awake at night.

But I also use my FF400 at 192 for measurements. Everything I've had before wasn't really up to that. The reason being that some interfaces don't even pass audio over 20 kHz, even if they announce 96 kHz sample rate. They seem to have fixed analog filters. And some others aren't very clean in what they pass at very high frequencies, but that depends on the environment. And you don't want noise you can't hear, do you?

I have to convert it later because the Behringer ADA8000 can only do 44.1/48 kHz and my MOTU 828 mkII is broken right now. Even still, the 828 mkII only has a single ADAT IN/OUT pair, so it can only accept 44.1/48 kHz ADAT. My MOTU 8PRE is capable of 88.2/96 kHz ADAT, but again, I'm using it with the ADA8000 so I'm stuck at 48 kHz.

I use REAPER to upsample the files via GLUE. I do so after all my edits and comping are complete, before I start mixing with FX or time-stretching.

I don't think any interface disables the analog LPF prior to ADC conversion. They may disable some filters if they're stacked, but I don't think that all LPFing gets disabled. They just don't need as steep of a filter at the higher rates as they do with 44.1 kHz. But, I suspect that most interfaces utilize the exact same analog LPF before the ADC regardless which samplerate you select.

Honestly, the LPF prior to the ADC may be what sets the high-end units apart from the budget interfaces. I don't know. They claim its their clock chip which makes the difference. I suppose the high-end gear uses more expensive clock crystals. I'm not in a listening environment accurate enough to tell any difference.

However, I think that recording at 96 is better than converting up to it, as my tests indicated to me that degradation of the audio occurs on the upsample. I would record at 96 kHz if I could.

[Bri1]

I suppose the high-end gear uses more expensive clock crystals.

However, I think that recording at 96 is better than converting up to it, as my tests indicated to me that degradation of the audio occurs on the upsample. I would record at 96 kHz if I could.

^indeed--the crystals do make them differences-- the crystals allow for better resonances from what i can gather..
i claim to not be a scientist or 'pro engineer' -<the only differences there sometimes is the paycheck--nothing else..

recording @96k is 2x better than 48k -that's common sense here....once you have the better q recordings--further manipulations can be downgraded to suit a faster,more computer friendly mixing flow..

if you rec @ 48 and do no processing...just a direct upsample 1:1 96k--there will be f all differences aside from sample amounts..#'s.
the only changes come if you process any new volumes or effects into your render out chains--- here you get them extra samples benefits of processes---but it won't alter the rec freq content unless affected to do so....pitch manipulations for eg.

@insub -bruv,while your @ reacomp,,try these settings with your tests if u cba >

[cyrano]

Originally Posted by insub

I have to convert it later because the Behringer ADA8000 can only do 44.1/48 kHz and my MOTU 828 mkII is broken right now. Even still, the 828 mkII only has a single ADAT IN/OUT pair, so it can only accept 44.1/48 kHz ADAT. My MOTU 8PRE is capable of 88.2/96 kHz ADAT, but again, I'm using it with the ADA8000 so I'm stuck at 48 kHz.

I use REAPER to upsample the files via GLUE. I do so after all my edits and comping are complete, before I start mixing with FX or time-stretching.

You're already doing it that way. Great!

EDIT: This GLUE?

https://cytomic.com/index.php?q=glue

Looks interesting.

I don't think any interface disables the analog LPF prior to ADC conversion. They may disable some filters if they're stacked, but I don't think that all LPFing gets disabled. They just don't need as steep of a filter at the higher rates as they do with 44.1 kHz. But, I suspect that most interfaces utilize the exact same analog LPF before the ADC regardless which samplerate you select.

I'm certain the RME FF400 does change filters. And I think the MOTU too. I''l have a look later, as I have one in atm that doesn't behave. I'll have to spank it, I think

Honestly, the LPF prior to the ADC may be what sets the high-end units apart from the budget interfaces. I don't know. They claim its their clock chip which makes the difference. I suppose the high-end gear uses more expensive clock crystals. I'm not in a listening environment accurate enough to tell any difference.

It's the clock too. But the differences aren't as big as they used to be. And it mostly shows (for me) when using two interfaces aggregated. If both are good, no problem. If one of them has a clock that's off, sync is lost after a while. In fact, that's my unscientific way of testing clock stability on old gear.

However, I think that recording at 96 is better than converting up to it, as my tests indicated to me that degradation of the audio occurs on the upsample. I would record at 96 kHz if I could.

Upsampling is VERY simple. It just adds "zero" samples. There shouldn't be any noise or artefacts. Of course, the LP filter following can produce artefacts. I wouldn't expect these to be audible in any way, though. But maybe I'm too confident in digital filtering?

[Bri1]

@insub > you could also try this chain @ 48k + 96k to see how it behaves...lol..
this chain works by raising the wet output on reacomp.. <it behaves different to 'normal uses'.. heh.be carefull.
then to test__> try raising either filter slowly (lp+hp) to check reasponses of each.k.
chain link> https://stash.reaper.fm/33518/STY%20...oosta.RfxChain

[insub]

Originally Posted by insub

Basically, I created a SINE sweep from 20 Hz - 10 kHz and automated the output gain of the tone generator to roughly output -3 dB/oct to mimic Pink Noise.

I sent a separate tone to ReaComp's sidechain input to maintain a -6 dB reduction with a 4:1 ratio (simulate mixing condition) because the GR would not remain static with the source signal. I set all other ReaComp settings to ZERO, including the RMS window.

At 0 ms RMS, the noise is everywhere in the spectrum and very clearly audible. But, even at 3-4 ms the noise is all but gone. My conclusion is that the default RMS window of 5 ms is not just to simulate an analog compressor (thought I read somewhere fastest hardware compressor circuits were approx. 5 ms reaction window) but rather to combat the extreme aliasing that occurs below 5 ms.

That doesn't mean there is no aliasing occurring. However the aliasing that remains is not solved with higher samplerate because it is folding back from the subharmonics extending below 0 Hz. These start folding back at frequencies at & below 1 kHz fundamental and were about -45 dB for the first subharmonic which eventually extends above the fundamental until it mates up with the first harmonic but doesn't appear to add together in amplitude with the first harmonic.

Additionally, I had never considered that subharmonics were generated from compression. In my test, ReaComp appears to generate equal subharmonics to the generated harmonics. Three harmonics in either direction, above & below the fundamental source tone. Each very low in amplitude. About -45 dB for the first, -70 dB for the 2nd, and I did not measure the level of the 3rd.

The GIF is very telling. I hope I can share it with you all soon. I tried PhotoBucket, but they do not allow GIFs to be uploaded.

Originally Posted by insub

I still cannot get the GIF to upload. So, here is the project file.
https://stash.reaper.fm/v/33516/Aliasing%20ReaComp.rpp

Unfortunately, you need Dead Duck's DD SigGen (FREE) for it to work because I did not know how to make a smooth, controllable (-3dB/oct) sweep with REAPER features. JS: Tone Generator changes musical pitches in steps, as far as I can tell. Get Dead Duck free FX pack here: http://deadducksoftware.blogspot.rs/

I think the file was WAYYY too big. I guess a static image is almost as good.

The sine sweep up has the RMS size set to 0 ms.
The sine sweep down has the RMS size set to 5 ms.


Looks like the file size limit may be 2 MB for the REAPER Stash.

[mschnell]

Originally Posted by insub

I set all other ReaComp settings to ZERO, including the RMS window.... But, even at 3-4 ms the noise is all but gone.

By any design, be it analog, digital or DSP, a "Compressor" with zero attack and release is not a compressor but a curve shaper. and that obviously needs to introduces harmonics.

-Michael

[pipelineaudio]

The rms window size ina really meaningful way IS the aliasing

You are creating a faster non linear process and therefore getting into more aliasing issues.

Watch mymaliasing video, over sampling does very very very little compared to fast non linear processes

[Bri1]

heh- all i would like is people on the same page getting the very best of the best.

if a newer reavised sampling system would not happen am totally happy to settle for a new single sample rate.. 60khz input-60khz output.
no upsampling--no downsampling-- just 1:1 in n out all along the production chain.
the less reasampling and filtering going on the better__ as that obviously causes 'issues'...some are subtle,some are not.
more sampling options are great,but if they are not used-they pretty much become useless..like anything.

^cannot see this happening though-- too many plugins,interfaces + sample libraries would need reawrites--but that does not mean all new stuff can't fit a newer 60khz srate!
onwards and upwards.

[jazznfunk]

Some (all?) RME interfaces (FF800 100%) offer a 64 kHz sample rate.

"Sample rate internally: 32, 44.1, 48, 64, 88.2 kHz, 96 kHz, 128, 176.4, 192 kHz"
(http://www.rme-audio.de/en/products/fireface_800.php)

But IMHO next to none plugins would be happy with that. And we must go up or down to standart set of samplerates anyway at the final production. So there's no benefit at all.

[Judders]

Originally Posted by clepsydrae

Interesting thread -- to summarize what's new vs. this thread I would say:

- Hugh Robjohns argues that modern anti-aliasing filters, though in theory they can be great, usually cut corners and thus in reality do still risk some fold-back distortion from the region just above nyquist

- several folks mention the potential value of high sample rates capturing more accurate arrival-time information and thus potentially representing the stereo field with more fidelity (i.e. it's not about frequencies, it's about localization)

I just posted in that thread seeking some more info.

There are several references in the AES paper (Reiss) that I linked before that address the localization question: Kunchur and Yoshikawa were two authors with papers on the subject (or related subjects).

It's seems like another thing that could be easily tested and have examples posted.

Please let us know if you get a reply!

[Bri1]

Some (all?) RME interfaces (FF800 100%) offer a 64 kHz sample rate.


And we must go up or down to standart set of samplerates anyway at the final production.

^not all interfaces offer 64khz -but @64khz ya know da skore!
i think some of the earlier libraries were even as low as 30,32khz samplings--but people still used them banks fine for years i guess...

6+4=10=1+0=computer happier.
+ we do not really need to go up or down with reasampling-- just 1 rate for lossless=simplez.. lossy could be 32khz wavpack as default lossy streaming and distributions.. =yeh.
who cares...

[insub]

Originally Posted by mschnell

By any design, be it analog, digital or DSP, a "Compressor" with zero attack and release is not a compressor but a curve shaper. and that obviously needs to introduces harmonics.

Potato - pohtahtoe.
That doesn't discount what I was trying to show.
Adjusting the Attack & Release times to 10 ms while the RMS size is still zero creates a similar, albeit less extreme, result. All of the harmonics past the first are -80 dB or below. The 1st harmonic & subharmonic are still about -50 dB.

Increasing the AA to 64x appears to have no affect on the generated harmonics or their aliases.

Originally Posted by pipelineaudio

The rms window size ina really meaningful way IS the aliasing

You are creating a faster non linear process and therefore getting into more aliasing issues.

Watch mymaliasing video, over sampling does very very very little compared to fast non linear processes

Link please.

[bezusheist]

Originally Posted by insub

Increasing the AA to 64x appears to have no affect on the generated harmonics or their aliases.

See here > https://forum.cockos.com/showthread.php?t=122197

[karbomusic]

Originally Posted by insub

Potato - pohtahtoe.
That doesn't discount what I was trying to show.

I'm sure I don't know what you guys are going on about per se (not very interested) but what he's saying is that if that sine wave changes in any way whatsoever to where it is no longer a perfect sine wave, harmonics are the result... Even if it were 100% in the analog domain.

[insub]

Originally Posted by bezusheist

See here > https://forum.cockos.com/showthread.php?t=122197

I think I remember reading that years ago and forgot all about it. I suppose that I assumed that something else had been fixed in this regard.
-----------------------

What I did find in ReaComp was that leaving the Attack & Release at 0 ms and increasing the RMS size completely eliminates (-80 dB) all harmonic distortion by 50 ms RMS size setting. This doesn't require any oversampling even at 48kHz samplerate.

The reality, however, is that even at 5 ms RMS size the harmonics are very low, around -40 dB compared to the signal, even at these extreme Attack & Release settings.
The colors in Spectrogram meter are about: Light Pink = -10dB, Magenta = -20dB, Red = -50dB, Yellow = -70dB, Green = -90dB, as best I can tell.


What I find most interesting is the aliasing of the subharmonic, which perhaps indicates an argument for performing harmonic generating processes in analog outside the box.

[pipelineaudio]

Originally Posted by insub

Potato - pohtahtoe.
That doesn't discount what I was trying to show.
Adjusting the Attack & Release times to 10 ms while the RMS size is still zero creates a similar, albeit less extreme, result. All of the harmonics past the first are -80 dB or below. The 1st harmonic & subharmonic are still about -50 dB.

Increasing the AA to 64x appears to have no affect on the generated harmonics or their aliases.


Link please.

https://www.youtube.com/watch?v=QCANuKa9ULU

[mschnell]

Originally Posted by insub

Potato - pohtahtoe.

Yep.
But with a curve shaper you intend to create harmonics/artifacts, while with a Compressor, you consider adding harmonics/artifacts as a necessary side-effect (if you in fact know that they necessary are created, even if very low intensity with not too short attack and release).

And regarding this discussion: the algorithm for curve shaping or compression does need to make sure, that the signal stays band limited, otherwise there will be additional unharmonic "fold back" distortions according to the Sample rate.

And in fact "RMS size" smooths the Volume control signal and hence effectively increases attack and release times.
http://forum.samplemodeling.com/sear...=active_topics
For really doing "compression", effective attack and release times below the lowest audio frequency's cycle time fed to to device does not make much sense, and other Volume-effects (i.e. multiplication, e.g. ring modulator) but compression or tremolo are supposed to create artifacts.

-Michael

[Bri1]

Heh--time to warm this up again..
there will only be another thread appearing later otherwise...confusions still in minds..
no reasponses or replies makes topics go cold-- this is a topic which would benefit from clarity of a majority view..?
did reacomp change effect when different s.rates are used in any 'tests'?



or is anybody actually bothered by any of this at all?
am guessing twatter or facebot gets more attention...lol-- nicer distractions.
i want to be treating this as a preciousss- 1 s.rate to rule them all..!

[superblonde.org]

Originally Posted by Judders

And here is an interesting straw poll showing that, in this limited sample pool, shows professional engineers work mostly at 44.1 kHz or 48 kHz:

http://www.recordingbootcamp.com/201...-the-pros-use/

Industry is entrenched in de facto standards & unable to upgrade due to interconnected systems



After reading this entire thread it's still not clear what sample rate I should set my Focusrite to. It goes up to 96k.

Mathematically it is better to choose multiples of rendering rate (44.1k final would choose either 44.1k or 88.2k for recording and 48k would choose either 48k or 96k recording). Not cross the numbers like record at 48k to render to 44.1k. Anything internet-video (i.e. for youtube) should then use either 48k or 96k because MP4 will prefer 48k or higher.

So...as for picking between 48k or 96k..... ??


Note, Behringer XR18 only supports: 24 bit at 44.1k or 48k with "signal processing 40-bit floating point"

[insub]

Recording at the higher rate can reduce aliasing for functions within the DAW that do not support/allow oversampling. Such as, time stretching and EQ cramping near Nyquist for ReaEQ.

Higher starting sample rate will require less oversampling for plugins/processes that need it.

There is also the possibility that your latency will be reduced at the higher sample rates due to the buffer getting filled faster by higher data audio streams.

Storage space is not an issue for most people with modern computers.

You will NOT hear any better fidelity of the RECORDED file by recording at higher sample rate than 44.1 kHz. MAYBE there will be some improved fidelity if your interface uses different ADC pre-filters depending on the selected recording sample rate. Most people don't know and most interface documentation doesn't specify if there are more than one ADC pre-filter.

I suggest recording at the highest rate your computer system will handle well. If 100 tracks of 96 kHz audio is killing your CPU, then you probably need to work with lower sample rates.

The SRC built into REAPER are very good. It should not make much difference between selecting 88.1 or 96 kHz and having to down-sample later.

XR18: 44.1 kHz or 48 kHz is plenty of fidelity. You can still benefit from SRC to a higher rate after recording and prior to mixing to get some of the benefits above.

People will argue the practical benefits of all the above advice, however they can be tested and quantified. Can one hear the difference by the end of a mix? You decide...

Of note: I think that the BluRay audio for video standard is 96 kHz (for playback system DSP benefit) and the latest pro audio interfaces all have 192 kHz recording rates, if that indicates anything.

[superblonde.org]

Originally Posted by insub

Of note: I think that the BluRay audio for video standard is 96 kHz (for playback system DSP benefit) and the latest pro audio interfaces all have 192 kHz recording rates

Blu-ray is 48 kHz, 96 kHz, or up to 192 kHz in some containers. I'd guess most discs are using 48kHz audio.

Neil Young however has recently produced and/or remastered much of his music in the highest bit depth and sampling rates, some of which can be streamed (as wav) thru his online music site. That could be useful for listening tests.


there is a difficulty with storage space: backing up the data in permanent form for longer term storage, which means "1+TB HD" or "256GB SD" is not a reliable medium, so with only 30GB per optical disc it does set some obstacles to recording with highest resolutions.

[batcat]

Originally Posted by superblonde.org

Blu-ray is 48 kHz, 96 kHz, or up to 192 kHz in some containers. I'd guess most discs are using 48kHz audio.

Neil Young however has recently produced and/or remastered much of his music in the highest bit depth and sampling rates, some of which can be streamed (as wav) thru his online music site. That could be useful for listening tests.


there is a difficulty with storage space: backing up the data in permanent form for longer term storage, which means "1+TB HD" or "256GB SD" is not a reliable medium, so with only 30GB per optical disc it does set some obstacles to recording with highest resolutions.

Neil Young is swimming deep in the audiophile swamp of blind belief.
"Why you don't need 24 Bit 192 kHz listening formats"
https://people.xiph.org/~xiphmont/demo/neil-young.html

That link is from the text info found in this great mythbusting video
D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org)
https://www.youtube.com/watch?v=cIQ9IXSUzuM

Also a good video
"High-Resolution Audio Demystified"
https://www.youtube.com/watch?v=Z5S_DI99wd8
Some are actually reselling their old stuff recorded with old gear and its limitations, in new Mega ultra bat-ear high-res sampling and bit format without telling
customers that those high res samplerates and bits can't just magically appear from nothing (the old recordings didn't capture it in the 1st place).
That's bad business ethics and fraud in my book.

[pipelineaudio]

Originally Posted by superblonde.org

Industry is entrenched in de facto standards & unable to upgrade due to interconnected systems

Nope, they're quite happy to spend their entire budget on moving to whatever is on the inside front cover of mix magazine this month. The BIG money is in hyperconformity so that's where its all targeted. I watch guys setup 192 khz systems in untreated rooms. And buy 300 dollar power cables

After reading this entire thread it's still not clear what sample rate I should set my Focusrite to. It goes up to 96k.

Its easy, if you are planning on making music for dogs and bats to listen to, record at 96khz

Mathematically it is better to choose multiples of rendering rate (44.1k final would choose either 44.1k or 88.2k for recording and 48k would choose either 48k or 96k recording). Not cross the numbers like record at 48k to render to 44.1k. Anything internet-video (i.e. for youtube) should then use either 48k or 96k because MP4 will prefer 48k or higher.

That certainly feels right, but its actually not true

[Judders]

Originally Posted by superblonde.org

Mathematically it is better to choose multiples of rendering rate (44.1k final would choose either 44.1k or 88.2k for recording and 48k would choose either 48k or 96k recording). Not cross the numbers like record at 48k to render to 44.1k. Anything internet-video (i.e. for youtube) should then use either 48k or 96k because MP4 will prefer 48k or higher.

Yeah, as Pipeline said, that's not true. At one point in time, a few decades ago it was the case for some SRC, but now they factor up to a common denominator so there's no error in the decimals.

The interpolation of REAPER's SRC is so good now that you don't have to worry about up or downsampling at all. It just shouldn't be a concern.

[emarsk]

Originally Posted by pipelineaudio

if you are planning on making music for dogs and bats to listen to, record at 96khz

On the other hand, if you make music for cats, 96kHz isn't enough: those cute little killers can hear up to 64kHz.

[Pashkuli]

Originally Posted by emarsk

On the other hand, if you make music for cats, 96kHz isn't enough: those cute little killers can hear up to 64kHz.

That is because they somehow had to develop or be created to hear the squeaks of rats and mice which are up in the ultra-zone.

24 bit/96kHz is good enough. anything above that is just a waste of space and resources in general. I use 20bit/48kHz (although the USB audio interface is advertised as 24bit, we all know they are actually not 24bit)

[pipelineaudio]

Originally Posted by emarsk

On the other hand, if you make music for cats, 96kHz isn't enough: those cute little killers can hear up to 64kHz.

No wonder they are so neurotic!

Every time they see their humans set up a stereo system they are thinking "you may as well paint the Sistine Chapel with a broom you idiots!"

[Jeffsounds]

Originally Posted by adXok

That is because they somehow had to develop or be created to hear the squeaks of rats and mice which are up in the ultra-zone.

True story. I'm doing a mixing session when suddenly I'm hearing this incredibly LOUD squealing. I'm thinking, "What the hell is that?" and I start searching across the panel to see if there is any indication of what track it is coming from and I see nothing. Everything looks normal but the squealing continues. So I stop the playback and I'm still hearing the squeal only now it sounds like it's coming behind me. I turn around and what do I see? Our cat has sneaked into the studio, where he is forbidden, and has found a mouse in the back among some boxes. He's holding the mouse in his mouth and the mouse is SHRIEKING to high heaven! I told him to let it go and he did and the mouse disappeared into the boxes and hasn't been seen since.

[superblonde.org]

The recent replies are essentially saying that the Focusrite (with extra $ price tag) with top end of 24-bit/192k is no better than the Behringer XR18 (with lower price tag) with top end of 24 bit/48k.

[karbomusic]

Originally Posted by adXok

24 bit/96kHz is good enough. anything above that is just a waste of space and resources in general. I use 20bit/48kHz (although the USB audio interface is advertised as 24bit, we all know they are actually not 24bit)

Pretty much, after 20 years of digital recording, I never pull up a project or a render from that 20 year spread and have any idea, or care what the original sample rate was. There are scenarios where 96k is beneficial but the vast, vast majority of the time there are a lot bigger fish to fry in other areas. I've been at 48k for about 10 years now because I don't typically use/do anything that needs more in my real-world situations.

[superblonde.org]

Real world situation: which combination is better listening quality?

Step 1:

A) Record & mix at 24-bit/96k.

B) Record & mix at 16-bit/48k.


The above comments suggest that they are "the same quality" therefore (B) is better (for data space and bandwidth).


Step 2:

C) Render into a video container with 16-bit PCM audio. Upload to youtube which converts the audio to AAC 128kbit.


D) Render into a video container with AAC (128 kbit) audio. Upload to youtube which unpacks the container's AAC audio and recontainerizes it without re-encoding.


The assumption based on recent replies would be that (either A or B)+(C) is a better end result. But perhaps there could be an argument made which says that (A)+(D) could be a better end result since the local AAC encoder might perform better with more resolution & more bits.

[cyrano]

Originally Posted by superblonde.org

The recent replies are essentially saying that the Focusrite (with extra $ price tag) with top end of 24-bit/192k is no better than the Behringer XR18 (with lower price tag) with top end of 24 bit/48k.

'Better" is a subjective qualification...

For music playback, all you really need is 16 bit, 44,1 kHz.

For audio editing, all you need is 24 bit, 44,1 kHz. Unless you are editing audio for video, in which case you need 24 bit, 48 kHz. Unless you're recording bats...

Etc.

Not everyone has the same needs. So the question "What's best for mixing audio?" is a bit besides the point, imho.

[superblonde.org]

Originally Posted by cyrano

For audio editing

Audio _editing_ can use 40-bit floating point (tho no one does). Did you mean to say audio _recording_ ?

[zeekat]

Originally Posted by superblonde.org

Audio _editing_ can use 40-bit floating point (tho no one does).

Technically, don't everyone here edit at 64bit float? It's Reaper internal default I think?