An Argument For Recording At High Res...

[karbomusic]

Here is another way to look at it. A average guitar can only put out about 1.2 volts peak (aka smacking a big power chord) give or take a couple hundred millivolts. So if we do the math on that we see it about the same as my quick test when using 1 millivolt to 1.2 volts of DR (which is likely larger than real life):



So back to the original point, using gain/compression on guitar recordings isn't really relevant because 16 bit can capture it all (especially what we actually need) without any compression anyway.

[superblonde.org]

"by turning the amp down and up"

that phrase makes me consider that the proposed experiment is not understood properly.

when I play my electric guitar thru my rig to my focusrite, I get noise floor of -60 dBr and my playing peaks at lets say -12 dBr to -6 dBr. That noise floor (and thus dynamic range) is a limitation of my entire signal chain and also at some point a limitation of the focusrite itself. Presumably I could improve my signal chain to lower the noise floor to maybe closer to -70 dBr. Presumably I could record a guitarist who has better playing technique and who would fully utilize musical articulation within that range of -60 .. -6. Could this playing range be broadened wider? Yes but only if the noisefloor could be reduced further because that is the limiting factor. (Would the softest note in the range be audible by the listener? well it depends if it is an expert listener.)

In this recording process the amp settings or recording levels are never touched.

None of this has anything to do with "being loud" because I simply turn down the monitoring volume, and if I want more volume, I turn the monitor up.

edit: it doesnt have anything to do with gain either. if the dynamic range isn't enough, then add more gain. if your gain is already maxed out, then you are using the wrong preamp, you need a preamp with more gain. gain = output/input. who cares if gain needs to be "1200" ? that's all gain is. turning up gain doesn't mean the guitar becomes distorted (altho rock n roll amps are designed that way). Gain only means you are amplifying the sound of the fingers hitting strings, and subsequently the dynamic breadth between hitting a single string as very very lightly as possible vs. hitting all strings as hard as possible.

[karbomusic]

What happens below the noise floor is always irrelevant until you remove the noise floor which you can't.

I get what you want to say, the only thing I'm pointing out is that it simply doesn't change anything bit rate wise if kids are using uber-distorted guitar tones these days, 16 bit alone could have captured everything larger than that (aka clean) anyway - even without a noise floor, the level would be so low (if you could achieve it), it would be useless (and still not 96 dB DR). My noise floor is around -75 dbFS in this particular preamp, I'm never going to have an audio signal that lives anywhere near that far down for example, much less living there and able to peak to zero dBFS.

And no, I'm not talking gain either - it's the simple fact that no guitar can produce 96 dB of DR in any usable fashion. Like I said, 96 dB is a lot, so much that we all tend to forget how much that actually is until we try to produce that range.

[superblonde.org]

Using a VST noise gate set at about -60 dBr followed by a VST expander in reaper would allow for recording wider dynamic range beyond the noise floor limitations of the analog signal chain. Also independent of guitar-tone-saturation-distortion, meaning, it could still be 'clean guitar'.

If done at 24-bit then the recording would allow fuller resolution of sound (because the signal is an amplified version of the volume range) compared to if the interface were set at 16-bit and recording done at 16-bit. Then later rendering to 16-bit (if desired) would still presumably be better than originally recording at only 16-bit.

[karbomusic]

Originally Posted by superblonde.org

Using a VST noise gate set at about -60 dBr followed by a VST expander in reaper would allow for recording wider dynamic range beyond the noise floor limitations of the analog signal chain. Also independent of guitar-tone-saturation-distortion, meaning, it could still be 'clean guitar'.

You should try it out and post the recorded result. I don't say that as a challenge, I say it because the only thing that really matters is what we can demonstrate to ourselves and/or others. I've been testing much/some of what I say/deduce over the last couple years and tbh, it's saved me an ass load of time because it removes unneeded worry about stuff that really doesn't matter. Except we can't know what matters until we try/test it ourselves.

And of course keeping practicality and real world in mind because that's the only thing we will ever actually use.

[cyrano]

Originally Posted by karbomusic

I mostly agree but that bombastic wildly varying snare hit will clip less at 24 bit and/or have more wiggle room for turning up after the fact.

My neighbour, who's playing Richard Clayderman around two o'clock at night, with the windows wide open, might agree. He plays like he hates that poor piano

[zeekat]

Found the thread with classical music fans complaining the CD releases are ridden with too big dynamic range

https://www.talkclassical.com/3155-d...k-dynamic.html

Didn't read past first page, but from that (and the survey up the page) the problem of 16 bit CDs being too good seems significant

[superblonde.org]

Originally Posted by zeekat

Found the thread with classical music fans complaining the CD releases are ridden with too big dynamic range

https://www.talkclassical.com/3155-d...k-dynamic.html

Didn't read past first page, but from that (and the survey up the page) the problem of 16 bit CDs being too good seems significant

well then here is the best part

I've run a number of my CDs through the TT Dynamic Range Meter Software (download). Interesting results and a remarkably wide spread. DR values from 5 to 8 are the norm for pop music releases nowadays. 10 to 15 seems to be the usual range for classical music.

26,6 Nono: Fragmente – Stille, An Diotima; LaSalle Quartett; 1986; DG
21,8 Nono: La lontananza nostalgica ... etc; Kremer/Kridenko; 1992; DG
18,1 Nono: Choral Works; SWR Vocal Ensemble; 2001; Hänssler
16,9 Pärt: Passio; Tonus Peregrinus; 2003; Naxos
16,4 Beethoven: Sinfonie 9; Nagano; 2011; Sony
15,9 Beethoven: Sinfonie 9; Herreweghe; 1999; Harmonia Mundi
15,6 Brahms/Strawinski: Violinkonzerte; Hahn; 2001; Sony
15,5 Schubert: Große C-Dur-Sinfonie/Unvollendete; Mackerras; 1998; Telarc
15,4 Barber: Sinfonien 1 & 2; Alsop, Marin; 2000; Naxos
15,1 Mendelssohn/Schostakowitsch: Violinkonzerte; Hahn; 2002; Sony
15,0 Beethoven: Sinfonien 3 & 8; Järvi; 2006; Sony
14,8 Schönberg/Monn: Kammersinfonie/Konzerte; Brogli-Sacher; 2013; Cybelle
14,7 Strauss: Ein Heldenleben/Macbeth; Markson; 1999; Naxos
14,6 Bruckner: Sinfonie 8; Furtwängler; 2005; Membran
14,6 Sibelius/Schönberg: Streichquartette; Tetzlaff Quartett; 2010; Avi
14,4 Barber: Cellokonzert etc; Alsop; 2001; Naxos
14,4 Beethoven: Piano Sonatas Vol. VIII; Schiff; 2008; ECM
14,2 Brahms: Sinfonie Nr. 1; Berglund; 2001; Ondine
14,2 Brahms: Streichquartette op. 51; Quartuor Ludwig; 1999; Naxos
14,1 Beethoven: Sinfonie 9; Karajan; 1977 [Remastered 2007]; DG
14,1 Beethoven: Sinfonien 5 & 6; Zinman; 1997; Arte Nova
14,0 Bruckner: Sinfonie 9; Davies; 2005; Arte Nova
14,0 Beethoven: Opp. 109-111; Leonskaja; 2010; MDG;
14,0 Strauss: Don Quichote etc; Markson; 2000; Naxos
14,0 Baird/Knapik et al.: Streichquartette; Quartetto Dafô; 2002; DUX
13,8 Beethoven: Sinfonie 9; Karajan; 1984; DG
13,8 Beethoven: Sinfonien 1 & 2; Karajan; 1985; DG
13,7 Poulenc: Stabat Mater etc; Järvi; 2013; DG
13,6 Beethoven: Op. 2; Kodama; 2008; PentaTone
13,5 Ligeti: Works for Piano; Aimard; 1996; Sony
13,5 Pärt: Spiegel im Spiegel; Hudson/Klinger/Kruse; 2006; Brilliant Classics
13,4 Beethoven: Sinfonie 8/Große Fuge; Nagano; 2011; Sony
13,3 Bruckner: Sinfonie 8; Boulez, Pierre; 2000; DG
13,3 Mahler: Sinfonie 1; Solti; 1964; Decca
13,2 Fauré: Requiem; Summerly; 1994; Naxos
13,1 Bruckner: Sinfonie 9; Furtwängler; 2005; Membran
13,1 Bruckner: Sinfonie 9; Jochum; 1982 [Remastered 2000]; EMI
13,1 Serocki/Baird/Krenz: Klavierkonzerte; Wodnicki; 2008; DUX
12,9 Górecki: Miserere; Gershon; 2012; Decca
12,9 Beethoven: Sinfonien 5 & 6; Mehta; 2009; Helicon Classics
12,8 Scattered Rhymes; Orlando Consort; 2008; Harmonia Mundi
12,8 Bruckner: Sinfonie 9; Furtwängler; 2001; Cantus Classics
12,7 Purcell: Instrumentalmusik; Hengelbrock; 1991; DHM
12,1 Beethoven: Sinfonie 9; Karajan; 1963; DG
12,1 Chopin: Etüden; Perahia; 2002; Sony
11,8 Bach: Kunst der Fuge; Aimard; 2008; DG
11,8 Schubert: Unvollendete/Große C-Dur-Sinfonie; Furtwängler; 2010; Naxos
11,6 Bach: Goldberg-Variationen; Gould; 1982; Columbia
11,5 Bach; Grimaud; 2008; DG
11,3 The Armed Man: A Mass for Peace; Jenkins; 2001; Virgin
11,2 Bruckner: Sinfonie 9; Giulini; 1989; DG
11,1 Bach: Französische Suiten 1-4; Gould; 1973; Columbia
10,8 Vivaldi – The Four Seasons; Richter; 2012; DG
10,7 Bruckner: Sinfonie 8; Furtwängler; 1964 [Remastered 1998]; EMI/Testament
10,5 Bach: Goldberg-Variationen; Gould; 1956 [Remastered 2002]; Columbia
10,5 Beethoven: Sinfonien 5 & 6; Karajan; 1963; DG
10,5 Strauss: Alpensinfonie; Karajan; 1981; DG
10,2 Bruckner: Sinfonie 3; Young; 2007; Oehms
10,1 Bruckner: Sinfonie 9; Rattle; 2012; Warner

I'll add some pop music releases, just for comparison:

14,8 Fleetwood Mac: Rumours; 1977 [1984]
13,1 Eric Clapton: Unplugged; 1992
12,3 Björk: Debut; 1993
11,6 Nirvana: Nevermind; 1991
10,6 R.E.M.: Automatic for the People; 1992
9,9 The Beatles: Sgt. Pepper; 1967 [2009 Remaster]
7,8 Daft Punk: Random Access Memories; 2013
7,2 Radiohead: OK Computer; 1997
5,0 Oasis: Be Here Now; 1997
4,3 Red Hot Chili Peppers: Californication; 1999

[emarsk]

If you put a noise gate in front of a guitar and your signal is quieter than the noise you're trying to block, then guess what? the noise gate will block the signal as well. The noise gate is irrelevant in the expander example, and the expander example is itself no different than a wide dynamic range generated by volume automation. But all of this is irrelevant when we take into consideration the listening environment.

Even if the guitar would be capable of more than 96dB of dynamic range - which is not - and even if the noise floor of the recording chain would be so low to allow for that range - which is not - the listening environment wouldn't allow for that range anyway.

30dB SPL is a hell of a quiet room, but let's say that you are so lucky to listen in such a dead quiet place: the quietest notes in the music would still be more than 30dB SPL or you wouldn't hear them, and if the loudest notes are 96dB higher than that, those are at more than 126dB SPL. That would be literally deafening: your ears would seriously risk permanent damage.

Now, in practice when you record you must have a generous headroom to keep away from the risk of clipping, so you usually never use all of the bits you have available. Let's say you waste 4 bits, or about 24dB: you still have 12 bits, that is about 72dB of dynamic range. That's still a lot (By the way, 12 bits are what the digital delays, where the signal had to pass before being printed on many vinyls, had).


If you think you need more headroom, then yeah, 24 bits could be useful, but realistically you're just wasting disk space. Unless you're doing something very wrong with setting the preamps, 16 bits are usually plenty enough.

[superblonde.org]

Originally Posted by emarsk

If you put a noise gate in front of a guitar and your signal is quieter than the noise you're trying to block, then guess what? the noise gate will block the signal as well. The noise gate is irrelevant in the expander example, and the expander example is itself no different than a wide dynamic range generated by volume automation.

i dont think you understood the purpose of the software noise gate (which is to eliminate the noise floor after the interface). the expander then maximizes the gated signal to the maximum range. It is different than a volume automation because the expander has gain (i.e. 1:1.5) whereas volume automation will be 1:1. The purpose is to expand the limited-range guitar signal of -60 dBr..-10 dBr to (lets say) -100 dBr .. 0 dBr, therefore multiplying the 50 dBr range to 100 dBr. If you only did volume automation (after eliminating the noise floor), then you'd be losing a large area of unused dynamic range. But perhaps you could think of the end result as "normalizing volume".

altho you could say, that a nice expander would do both gate & amplification in one pass, without an additional noise gate on the front.

[Judders]

Just saw this today, and thought it might be of interest considering the dynamic range conversation going on:

The MixPre II recorders feature Sound Devices’ own patented topology of multiple analog-to-digital converters. These converters appear in the circuitry after the Kashmir microphone preamplifier stage. They enable the recording of very-low distortion, ultra-high dynamic range audio. The MixPre II’s A-to-D’s can resolve more than 142 dB of dynamic range. Together with their Kashmir microphone preamplifiers and 32-bit float files, the MixPre II captures audio that is limited only by the capabilities of the microphone.

https://www.newsshooter.com/2019/08/...oat-recording/

[karbomusic]

I think ^that's the one Cyrano mentioned in addition to Zoom.

[Judders]

Originally Posted by karbomusic

I think ^that's the one Cyrano mentioned in addition to Zoom.

Maybe worth mentioning it's a field recorder. Completely unnecessary for guitars!

[karbomusic]

Originally Posted by Judders

Maybe worth mentioning it's a field recorder. Completely unnecessary for guitars!

Earth has a fairly wide DR.

[emarsk]

Originally Posted by superblonde.org

i dont think you understood the purpose of the software noise gate (which is to eliminate the noise floor after the interface).

Yes, I know how a noise gate works. The signal it sees is noise floor + guitar signal; if you set it to open at a threshold, the guitar signal needs to be higher than that threshold or it gets blocked. And when it's above the threshold, the whole signal passes, guitar and noise floor together.

Unless you mean a denoiser, which is a completely different beast, and really not recommended as a recording tool.

An expander gets the loud notes louder and the quiet notes quieter, it's just the reverse of a compressor. My point is, you can expand the dynamic range in the mix with expanders, volume automation, arrangement and layering; the example wouldn't change in the least, it's irrelevant either way.

[superblonde.org]

I guess the compromise is simple: record everything at 24-bit/96 kHz into 256kBit OGG Opus Quality 10 Hard-CBR independent-channel.

edit- oops, Opus only supports 48 kHz max as inputs.

[Icchan]

I'll just drop this here... made this some years ago already but it fits again:

[emarsk]

Originally Posted by Icchan

I'll just drop this here...

To be honest, I don't see a great deal of fighting here, we're mostly nitpicking for the fun of it, aren't we?

[Jorgen]

Originally Posted by cyrano

My neighbour, who's playing Richard Clayderman around two o'clock at night, with the windows wide open, might agree. He plays like he hates that poor piano

Oh man, was that a blast from the past!

Anyways, I experimented recently with recording a song in 96/24, mostly virtual instruments and vocals. To begin with, I found the synth sounds to be too huge so I had to work on that a bit. I did some vocals today and the whole thing might be finished. To my ears, it's the best sounding production I've ever made, but then again it's a ballad with lots of ear candy and strings and the whole shebang. Fact is it's almost too pretty, but maybe that's a good thing in this particular case.

[Icchan]

I just don't have the energy to read the whole thread, but there's really good reason to record 24bit. And more complex reason to use higher sampling rates than 44.1,but not the reason you might expect.

About 24bits:

Unless you're able to always maximize your signal to noise ratio when recording, it'll help you to prevent "noise compounding" where each successive track you add to your mix has some noise and it all gets summed together in the end and can compound into a problem.

Of course, in digital domain we have 32 or 64bit engines in the DAW's that allow you to turn the signal down and the noise doesn't just stay in the noise floor of 24bit signal but goes lower as well with the signal, so that "problem" isn't really that relevant anymore.

But regardless, it gives your recordings lower signal to noise ratio (in the confines of your preamp and signal chain in general) so there's no real reason not to use it. WHEN RECORDING AND MIXING... for the reproduction, noise level of 16bit is way, way enough. And 24bit doesn't have any use for listening music or other auditory material.

Most devices can't even have noise low enough to even use that 24bits effectively... most get to 22bit mark or so before their inherent noise takes the rest of the bits so... don't worry about it!

As for the sampling rate...

Please for the love of everything that's good in the world, study Nyquist sampling theorem PROPERLY AND FULLY (might take some years of studying maths and physics before you can understand Nyquist), it clearly states, that:

Sampling frequency has to be twice the frequency of the highest frequency you want to reproduce, and then you're able to fully reproduce the signal as it was. PERIOID!
In case of 20khz hearing limit of humans, it would be 40khz..

But people don't read everything the theory says, and they skip phenomena known as ALIASING!

This is an issue, where any frequencies in the SAMPLED signal that are higher than your Nyquist limit fold back (alias) below the Nyquist limit (in this example case 20khz). And it's reverse, meaning that 21khz frequency would fold back to 19khz and sum with it, 22khz would fold back to 18khz and sum... etc.

This is why we need anti aliasing filter, which is nothing more than LP-filter of certain Q (how steep it is) centered on the Nyquist limit (20khz) that's BEFORE your sampling stage. This tries to remove any signals that have higher frequency than your Nyquist and thus limit the amplitude of those aliasing artifacts.

This filter, like any other isn't perfect of course but has a slope. Quite common is 12dB or 24dB/Octave filter since they don't have such a pronounced hump before the -3dB point (completely separate topic regarding filters).

But issue is, that for example, if I have 20khz Nyquist limit (highest frequency I can or want to sample) and my AA-filter is 12dB/Oct. A frequency of 30khz would fold back to a frequency of 10khz, which is very audible as a tone, not just some fizzle, and would be only 12dB lower then it was before it was sampled. Suddenly you could get very audible artifacts from previously unheard signals from your surroundings (they were ultrasound frequencies before they aliased to hearing range).

This is of course only if your source (microphone or other device) is even capable of generating (or capturing) those frequencies and your pre-amp stage is able to even pass them on to be sampled in the first place. Usually all devices have their bandwidth limits at around 20khz or there about so you'd rarely get e.g 30khz tone all the way from the mic to the AD-converter.

Let's say that for some reason you did get significant ultrasound frequencies captured by your setup, what can you do? Well, why not sample at a higher rate than 40khz, let say 44.1khz!. Now you can start your AA-filter at 20khz, and by the time your signal passes that 22.05khz frequency, it has already been attenuated somewhat (~3dB?) already since an octave would be 20khz, and a signal folding back would have to be 2khz higher before it folds back. So it wouldn't be so severe... and anything higher than that would be attenuated even more before it folds back.

But ah, I hear you say... 3dB isn't much! And no, it isn't. So we can either make the filter steeper than 12dB/Oct, which could induce a resonant hump at the corner frequency, or we could INCREASE THE SAMPLING RATE! Because then any ultrasound frequencies have more time to attenuate before they fold back, or we could have less steep AA-filter and reduce the resonant peak at the corner frequency!!!

And there it is... that's why you might want to sample at higher sampling frequencies than let's say 44.1khz... to prevent ultrasounds from causing aliasing in your wanted signal.

But there's no free lunch, and a capture device that samples at higher frequencies needs to be properly designed from electronics perspective.

If the AA-filter isn't well designed or high sampling rate causes issues with the noise floor, you're not going to get the benefit of the higher sampling rates.

But in the end, the 44.1khz render of the final song or mix is perfectly adaquate in reproducing everything _unless_ you were stupid and had ultrasound signals in your nice 96khz master project... those will fold back when downsampling since they are above the nyquist unless you have AA-filter in your master bus to get rid of them, or your re-sampler doesn't have it built in... as it should have.

But in the end kids... learn your sampling theorem before fighting about sampling rates... they're not there why you might think they are.

[jlgrimes11]

Originally Posted by Jeffsounds

Makes sense to me...

High-Res... Why Not?
There is no argument for dumbing-down a recording, while recording

I know my hearing only goes up to about 15khz, so it would be worthless for me.

I know that some plugins sound better at high sample rates but I'd rather just work with better plugins that can oversample if needed and occasionally I know some plugins can sound worse at high sampling rates.

I would think most plugins are probably fine tuned to sound best at 44.1 or 48.

I remember working with a Reason effect at 88.2 and it lost its mojo up there.

[Judders]

Originally Posted by jlgrimes11

I know my hearing only goes up to about 15khz, so it would be worthless for me.

Aliasing folds back into the audible range.

Originally Posted by jlgrimes11

I know that some plugins sound better at high sample rates but I'd rather just work with better plugins that can oversample if needed and occasionally I know some plugins can sound worse at high sampling rates.

Oversampling can bring its own problems, and if you're running a lot of oversampled plugins you would save CPU by simply running your session at a higher sample rate.

Originally Posted by jlgrimes11

I would think most plugins are probably fine tuned to sound best at 44.1 or 48.

AES producer standards suggest 96 kHz, so it would be an odd choice for a plugin manufacturer if their plugins sounded worse at professional standards.

At the end of the day, if you can't hear any benefit and you're not taxing your system with a ton of oversampling, then it really doesn't matter. Run your session at whatever sample rate you like.

It's one of those endless arguments because there is no right or wrong answer, no matter how angrily some people try to shout that there is.

[cyrano]

It's real simple...

24 bit. Yes. Always.

96 kHz or higher? Not as clear cut. Usually, there's no need. But, some rare cases might need it. Some plugins can do better with higher sample rates, but these would be the exceptions.

I'm currently working on data from audio measurement systems. Tests performed by audiologists.

Nobody hears anything over 17 kHz. Exceptions exist, but are extremely rare. What's far more interesting, is hearing loss in young people. Around 40% (+/- 10%) have a hearing loss of about 8 dB. For the first time in man's history, some of the elder (>45 years old) hear better than the young. All audiologists agree on the causes. Number one is what you'd expect: headphones. But number two is a bit of a surprise: background noise. It seems background noise has risen from a natural 32 dB to between 40 and 60 dB, depending on where you live.

We're all going deaf. And it's probably happening to our other senses too. Too much staring at screens, too much sugar in our food/drink.

This renders a discussion about higher sample rates moot.

[Judders]

Originally Posted by cyrano

It's real simple...

24 bit. Yes. Always.

96 kHz or higher? Not as clear cut. Usually, there's no need. But, some rare cases might need it. Some plugins can do better with higher sample rates, but these would be the exceptions.

I'm currently working on data from audio measurement systems. Tests performed by audiologists.

Nobody hears anything over 17 kHz. Exceptions exist, but are extremely rare. What's far more interesting, is hearing loss in young people. Around 40% (+/- 10%) have a hearing loss of about 8 dB. For the first time in man's history, some of the elder (>45 years old) hear better than the young. All audiologists agree on the causes. Number one is what you'd expect: headphones. But number two is a bit of a surprise: background noise. It seems background noise has risen from a natural 32 dB to between 40 and 60 dB, depending on where you live.

We're all going deaf. And it's probably happening to our other senses too. Too much staring at screens, too much sugar in our food/drink.

This renders a discussion about higher sample rates moot.

For recording, yes.

For mixing with non-linear digital processing; the problem is inharmonic content entering the audible range. I would not say that plugins with non-linear processes are the exception at all.

The argument for saving CPU makes no sense if you are up and down sampling within plugins several times per track.

[mlprod]

Originally Posted by Icchan

I just don't have the energy to read the whole thread, but there's really good reason to record 24bit. And more complex reason to use higher sampling rates than 44.1,but not the reason you might expect.

About 24bits:

Unless you're able to always maximize your signal to noise ratio when recording, it'll help you to prevent "noise compounding" where each successive track you add to your mix has some noise and it all gets summed together in the end and can compound into a problem.

Of course, in digital domain we have 32 or 64bit engines in the DAW's that allow you to turn the signal down and the noise doesn't just stay in the noise floor of 24bit signal but goes lower as well with the signal, so that "problem" isn't really that relevant anymore.

But regardless, it gives your recordings lower signal to noise ratio (in the confines of your preamp and signal chain in general) so there's no real reason not to use it. WHEN RECORDING AND MIXING... for the reproduction, noise level of 16bit is way, way enough. And 24bit doesn't have any use for listening music or other auditory material.

Most devices can't even have noise low enough to even use that 24bits effectively... most get to 22bit mark or so before their inherent noise takes the rest of the bits so... don't worry about it!

As for the sampling rate...

Please for the love of everything that's good in the world, study Nyquist sampling theorem PROPERLY AND FULLY (might take some years of studying maths and physics before you can understand Nyquist), it clearly states, that:

Sampling frequency has to be twice the frequency of the highest frequency you want to reproduce, and then you're able to fully reproduce the signal as it was. PERIOID!
In case of 20khz hearing limit of humans, it would be 40khz..

But people don't read everything the theory says, and they skip phenomena known as ALIASING!

This is an issue, where any frequencies in the SAMPLED signal that are higher than your Nyquist limit fold back (alias) below the Nyquist limit (in this example case 20khz). And it's reverse, meaning that 21khz frequency would fold back to 19khz and sum with it, 22khz would fold back to 18khz and sum... etc.

This is why we need anti aliasing filter, which is nothing more than LP-filter of certain Q (how steep it is) centered on the Nyquist limit (20khz) that's BEFORE your sampling stage. This tries to remove any signals that have higher frequency than your Nyquist and thus limit the amplitude of those aliasing artifacts.

This filter, like any other isn't perfect of course but has a slope. Quite common is 12dB or 24dB/Octave filter since they don't have such a pronounced hump before the -3dB point (completely separate topic regarding filters).

But issue is, that for example, if I have 20khz Nyquist limit (highest frequency I can or want to sample) and my AA-filter is 12dB/Oct. A frequency of 30khz would fold back to a frequency of 10khz, which is very audible as a tone, not just some fizzle, and would be only 12dB lower then it was before it was sampled. Suddenly you could get very audible artifacts from previously unheard signals from your surroundings (they were ultrasound frequencies before they aliased to hearing range).

This is of course only if your source (microphone or other device) is even capable of generating (or capturing) those frequencies and your pre-amp stage is able to even pass them on to be sampled in the first place. Usually all devices have their bandwidth limits at around 20khz or there about so you'd rarely get e.g 30khz tone all the way from the mic to the AD-converter.

Let's say that for some reason you did get significant ultrasound frequencies captured by your setup, what can you do? Well, why not sample at a higher rate than 40khz, let say 44.1khz!. Now you can start your AA-filter at 20khz, and by the time your signal passes that 22.05khz frequency, it has already been attenuated somewhat (~3dB?) already since an octave would be 20khz, and a signal folding back would have to be 2khz higher before it folds back. So it wouldn't be so severe... and anything higher than that would be attenuated even more before it folds back.

But ah, I hear you say... 3dB isn't much! And no, it isn't. So we can either make the filter steeper than 12dB/Oct, which could induce a resonant hump at the corner frequency, or we could INCREASE THE SAMPLING RATE! Because then any ultrasound frequencies have more time to attenuate before they fold back, or we could have less steep AA-filter and reduce the resonant peak at the corner frequency!!!

And there it is... that's why you might want to sample at higher sampling frequencies than let's say 44.1khz... to prevent ultrasounds from causing aliasing in your wanted signal.

But there's no free lunch, and a capture device that samples at higher frequencies needs to be properly designed from electronics perspective.

If the AA-filter isn't well designed or high sampling rate causes issues with the noise floor, you're not going to get the benefit of the higher sampling rates.

But in the end, the 44.1khz render of the final song or mix is perfectly adaquate in reproducing everything _unless_ you were stupid and had ultrasound signals in your nice 96khz master project... those will fold back when downsampling since they are above the nyquist unless you have AA-filter in your master bus to get rid of them, or your re-sampler doesn't have it built in... as it should have.

But in the end kids... learn your sampling theorem before fighting about sampling rates... they're not there why you might think they are.

This is good summary.
If one feels like 44,1 is enough that is completely valid and music will happen anyway. But there can be some benefits with 2x sample rates in the processing stage mainly, not to mention that some labels want hi-res deliverables which means that for me in the mastering stage there is no reason not to use 96k in this day and age.

[serr]

Originally Posted by jlgrimes11

I know my hearing only goes up to about 15khz, so it would be worthless for me.

I know that some plugins sound better at high sample rates but I'd rather just work with better plugins that can oversample if needed and occasionally I know some plugins can sound worse at high sampling rates.

I would think most plugins are probably fine tuned to sound best at 44.1 or 48.

I remember working with a Reason effect at 88.2 and it lost its mojo up there.

HD sample rates have nothing to do with preserving frequencies above the range of hearing. It's all about not having to make an analog eq to filter out a sample rate that is right next to the high end of the audio data as happens in SD sample rates. That's a difficult steep eq to make and it turns out it affects audio in the range of hearing when it isn't done well.

Then there are additional issues like aliasing when converting HD audio to SD containers.

None of this has anything to do with any artifact above the range of hearing. (Above the range of what any speakers produce as well. You're not hearing anything from this no matter who you are or what ears you have.)

Plugins are going to be optimized to work with audio in the range of hearing. If they have some issue with one sample rate vs another... Actual bugs aside, aliasing is a technical consideration in this territory.


Maybe think of it like tape bias if that helps?
The tape bias frequency just whistles away unfiltered because it's above 40kHz. Neither speakers or ears will hear it.

[drumphil]

Originally Posted by serr

HD sample rates have nothing to do with preserving frequencies above the range of hearing.

Well, there are some edge cases where it is to do with preserving frequencies above the range of hearing.

Like with this paper that provides pitch shifted audio examples of the ultrasonic songs of male mice.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1275525/

Much like how space agencies transform information in the infrared and x-ray spectrum into the visual spectrum so that humans can see pictures and interpret information within the limitations of our senses.


As far as audio production goes, pitch shifting down may allow the creation of some interesting sounds from frequency ranges we can't ordinarily hear. I've heard that this has been used in movie sound effects, but I'm not sure how often it is actually useful.

Regardless of those edge cases, I think Dan Lavry had it right when he suggested adopting a 60K sample rate and calling it a day.

The tape bias frequency just whistles away unfiltered because it's above 40kHz. Neither speakers or ears will hear it.

That is mostly correct:

https://people.xiph.org/~xiphmont/demo/neil-young.html

Neither audio transducers nor power amplifiers are free of distortion, and distortion tends to increase rapidly at the lowest and highest frequencies. If the same transducer reproduces ultrasonics along with audible content, any nonlinearity will shift some of the ultrasonic content down into the audible range as an uncontrolled spray of intermodulation distortion products covering the entire audible spectrum. Nonlinearity in a power amplifier will produce the same effect. The effect is very slight, but listening tests have confirmed that both effects can be audible.

That's a difficult steep eq to make and it turns out it affects audio in the range of hearing when it isn't done well.

That is also correct, but given that the artifacts generated tend to be at nyquist frequency, which at standard sample rates puts it where the ears are the least sensitive, the importance is often overblown.

Anyway, not trying to argue with you as what you say is correct. I just think it is important to maintain perspective on the relative importance of these things.

Given the choice between an excellent studio, with nice sounding rooms, microphones, and analog stage electronics with good quality 44.1K 16bit converters, or an average studio with 96K 24bit converters, I'll take the nice studio any day.

Once you get to good quality converters at 44.1K 16bit, higher sample rates and bit depths are about the least important thing to getting better quality recordings in 99.99% of studios.

If someone told me that 44.1K 16bit converters were all humanity would ever have access to, I'd be like "yeah, righto, recordings are going to sound great", and just get on with it.

I can't come up with an example from my life where I've been in a studio where if someone said to me "hey, we can offer you a better studio, but you'll only be able to record at 44.1K 16bit" where I wouldn't have taken the offer.

[Judders]

Originally Posted by drumphil

Once you get to good quality converters at 44.1K 16bit, higher sample rates and bit depths are about the least important thing to getting better quality recordings in 99.99% of studios.

If someone told me that 44.1K 16bit converters were all humanity would ever have access to, I'd be like "yeah, righto, recordings are going to sound great", and just get on with it.

I can't come up with an example from my life where I've been in a studio where if someone said to me "hey, we can offer you a better studio, but you'll only be able to record at 44.1K 16bit" where I wouldn't have taken the offer.

I got on fine when 44.1 kHz 16 bit was all I had to record with.

But, as I said, recording isn't the issue.

I posted an example of a software bass synth going through an amp sim in a thread around here somewhere. The difference between 44.1 kHz and 96 kHz was not subtle.

It all just depends what you're doing, processing wise. Like I said a few posts ago; there is no right or wrong answer.

[Pashkuli]

Originally Posted by ernzo

Higher Bit Depth = BETTER (24bit < 32bit < 32bit-FP < 64?)

Higher Sample Rate = Pointless for most applications.

-Now, to debunk the debunkers and myself:
https://www.youtube.com/watch?v=4eC6L3_k_48

Here we can see how in fact a Vynil record extends nicely up to 96khz in most cases,
and even up to 120khz in the Brass secion of the second track (4:12)..

So yes, vinyls can extend beyond the 96khz mark,
and it could even be argued that all the extra information brings naturality to the sound.

Hey, ernzo
I am really interested in those sound experiments but my questions regarding "vinyls can extend beyond the 96khz" are:

1. What are those microphones that can record cymbals and brass harmonics "beyond 96kHz"... Ok, 96kHz is in digital domain, so 48kHz in acoustic sound?
With today’s scientific microphones it is possible to cover a frequency range starting from around 1 Hz and
reaching up to 140 kHz. I doubt those high freq. were used for music studio recordings... and then the mastering sent to vinyl manufacturer. Because they actually have very low sensitivity (need to be placed exactly at the source, usually those mics are not thicker than a big steel nail).

Vinyl technology is very unreliable and some experimental and specifically design vinyl masters could reach 50kHz and above.

2. How are we supposed to hear anything beyond the extreme upper range of human hearing 20kHz?

3. Who recorded/produced those 20kHz and above frequencies on that vinyl experiment video and how: microphones, analogue equipment, mastering?

4. Is it one of those specifically designed vinyls? Otherwise they seem like a regular record store vinyls.

I really would like to know.

[serr]

Right on @drumphil.
Sounds like we're on the same page. I don't mean to be stubbornly dismissive with that either. Edge cases, sound bending with pitch shifting are all well and good. Just trying to throw something out there for the folks stuck on wondering what they're missing. There are some things to talk about but you're not missing any audible audio at SD. 16 bit can actually be heard with the right recordings but I'll say it's still fair to call an edge case in the long and short of things.

I'm still blaming the across the board poor mastering work (or "portable device mastering" if someone believes this is intentional) that came to be with the CD format. When you hear this with every last CD you come across for years and then you hear examples of old vinyl pressings of the same recordings being cleaner, you just might start to think something is wrong with the 16/44.1 format. This has to be responsible for all this!

@Judders
That's a really good example of why I like to just set the switch to 96k and move on. If you had set to 96k and never checked up on it... turns out it would have been fine. But if you had set to 44.1k and never checked up on things...

I've never heard an edge case with 24 bit at 96k. Kind of a "We're done here." format.

[Pashkuli]

ernzo,
can you please see thеsе videoс?

https://www.youtube.com/watch?v=1dGU79-1DYM

https://www.youtube.com/watch?v=y_OUkLbSs24


How a vinyl can reproduce frequencies that have not been there in the first place? Or can it be that the cutting tool (that made the master vinyl) could reproduce frequencies around 48-50kHz?
This seems quite impossible to me as there is an interaction between the cutter and the soft plastic (vinyl) of the master vinyl.

DO you think that lacquering the master vinyl with silver is tiny enough to fill the 48 000Hz grooves or even 20 000Hz? Quite impossible, in my opinion.