Record at 48kHz Mix down 96khz

[azslow3]

I may be see it wrong... But if everything in a project is 48kHz and there is just "saving" into 96kHz... I expect all frequencies above 24kHz should be strictly zero... no? ?

I mean it is like using 64 bit depth for recording (from 24 fixed point interface > maximum resolution of any interface), does not make sense at all.

[Michal Ochedowski]

1 hour ago, azslow3 said:

I may be see it wrong... But if everything in a project is 48kHz and there is just "saving" into 96kHz... I expect all frequencies above 24kHz should be strictly zero... no?

That's correct. 

1 hour ago, azslow3 said:

I mean it is like using 64 bit depth for recording (from 24 fixed point interface > maximum resolution of any interface), does not make sense at all.

Again correct. It doesn't make sense. However example of Ableton Live comes to mind. Everything in this DAW is processed internally at 32 bit. As far as I'm aware, this behaviour can't be modified. So to avoid unnecessary bit depth changes, it might be a good idea to keep all internal recordings at 32 bit. It's even recommended in user manual.

On the other hand, in Reaper there is a field named: Track mixing bit depth. It gives user a choice, which is a great feature. 

Edited January 21, 2020 by Michal Ochedowski

[Starship Krupa]

On 1/20/2020 at 3:40 AM, Gswitz said:

Using all available bits (loudest moment in the song hits 0), 44.1 can theoretically reproduce everything a human can hear up to around 108 decibels at 16 bit.

(this rantlet is not directed at you personally, G, I just sound off about higher rates every once in a while)

This assumes of course that all there is to being able to measure what all humans are able to perceive with our ears is frequency response and that there is nothing more to learn about human hearing and sound perception and reproduction.

Sure, 95% of people can't tell the difference between a heinous low bitrate MP3 and a pristine FLAC, but for 99% of people, "Las Meninas" is also a picture of a bunch of history-looking people standing around in a bedroom. Sayin' you might have to know what to look/listen for, and that hearing perception and acuity can possibly be learned.

I say "possibly" because I am not 100% sure and I haven't performed rigorous double-blind tests. I will say that I hear the difference between 125 MP3's and FLAC's being played back in MusicBee but I failed the snot out of that NPR test thing. What it told me was that I can't tell the difference between those formats when they are delivered through a web browser. Gimme the same files and a bit-perfect music player and then see how I do.

There is a wide variation in color perception and even visual acuity (look at how many people wear eyeglasses) in the human population. 4% of music students possess the innate ability to identify absolute pitch while the rest of us have to learn relative pitch. Maybe there is more to hearing perception than can be measured with a frequency counter. Why can I, at age 58 after playing in loud rock bands for years, still dig a tiny finger squeak in a guitar track out of a dense mix?

I'm not saying it's so, I'm saying that I'm open to the possibility that there may be more to it than raw frequency response, and my own empirical observations suggest that there's a good chance. There may not be information up there that we can hear (and that our primitive paper-or-plastic drivers can even reproduce), but recording with the extra bandwidth may have an effect on things other than just frequency response, like phase or group delay of higher frequencies.

I personally still track at 44.1, but if I were running a pro studio, and had the disk space, I might do more at 88.2 or 96 just because we can. It will give future generations more to work with if they ever dig up what we do and want to work with it.

And, BTW, for the OP, the thing to do if you intend to mix down at 96 is to start the project at that rate. Otherwise, at least with Cakewalk, there is little point. Any advantage as far as plug-in sampling is already covered elsewhere in the program.

[Starship Krupa]

On 1/19/2020 at 7:41 AM, Byron Dickens said:

no delivery format is 96k

HDTracks and its fellow HQ download sites beg to differ.

I bought Radiohead's A Moon-Shaped Pool from them because it was the only downloadable way to get it lossless and oh man does that record sound good in 48K.

They have other albums available for purchase in 96K.

I don't think I would buy in that format myself, as I find that my limit of quality perception goes to "lossless" and that's about it. 256K AAC is pretty good, but I like my music like it was made, lossless.

[bitflipper]

Yes, there are music sites that offer 96 KHz and even 192 KHz wave files - for a premium price.

It's a scam that unethically takes advantage of consumer ignorance. Given that the material was in all likelihood simply upsampled from the original source, whatever perceived benefit from the higher sample rate couldn't possibly be in there. You're paying for the placebo effect.

[Gswitz]

I think 48 is more than placebo myself because the anti aliasing filter can start higher above 20k and be more gradual. But what evs. Bit, i know you were talking about double rates.

I do often track at double rates. Makes sense to me and gives me piece of mind.

I'm ok with the idea that maybe we can detect frequencies above 20k. So, to preserve those, you would have to track at double rates, mix stems and master at double rates and distribute at double rates.

The bands i record never ask me the rate i track at. But if they thought there was $5 in it, they'd ask for a double rate export. He he. (By this i mean, pop open that pricy download and look for content over 20. See any?)

If done correctly, a 96k file can have content up above 40k. See dog whistle. Quad rates required for a proper cat whistle ?. 

Edited January 21, 2020 by Gswitz

[AB99]

I track at 96K or 88.1 because of the latency benefits and frankly, the hard drive space, taxing the system, etc. is no longer an issue.  I mix to various sample rates - because as long as I have the mix where I want it, I may as well make extra copies at various sample rates.  

[Gswitz]

1 minute ago, Alan Bachman said:

the hard drive space, taxing the system, etc. is no longer an issue. 

I mostly agree, unless you are letting each instrument record unlimited takes and planning to keep them all in the project until comping. I hit my head on that ceiling still and my pc isn't shabby.

[Starship Krupa]

6 minutes ago, bitflipper said:

Given that the material was in all likelihood simply upsampled from the original source, whatever perceived benefit from the higher sample rate couldn't possibly be in there.

I've not looked into their descriptions, but yeah, it sure would be nice to know whether the record in question was actually mastered at that rate. ?Otherwise, as you say, even if the buyer does think that there's some benefit to be gained by playing a higher bit rate file back through their fancy DAC, they could get the same effect by running their 44.1K or 48K lossless file through a good resampling program.

And when I say "good" resampling program, I only wish that every audio program could up and down sample audio at every rate without bungling it. The fact that not all of them do is an empirically proven fact. Our dear Cakewalk's resampling algorithms handle the job in pristine fashion.

(There's a site that shows test results on this, and I've resampled sine waves in upward and downward directions in programs I use then compared the results using SPAN. Oh man.)

I was surprised that the Radiohead album was offered lossless at 48K and not at 44.1K. That implies to me that they're working at 48 rather than the old CD-friendly 44.1. I still work at 44.1 because....habit? Tradition? My Firepod takes a moment to switch rates when I play back different files.

[Gswitz]

I think the gentler filter is the best argument for 48 imho.

For those who don't know, if you play back digital audio with content above the nyquist frequency you get artifacts. That is partly why 44.1 was originally chosen. Half of 44.1 is 22.05. To avoid aliasing noise on reproduction, a sharp low-pass filter removes content above 20k to nothing by 22.05k, just above the scientifically asserted upper edge of human hearing. So, when you playback the file, there is nothing at all above the nyquist frequency. This sharp eq has its own impacts.

Doing the same between 20k and 24k is a more shallow filter and less discernable. The 24k ceiling is for the 48k rate. At double and quad rates it hardly matters where the filter starts as long as there is one. If you don't use the filter, the aliasing is clearly in the audible range double rate or no.

You do not apply this filter. Your interface and daw do it. The quality of this filter and how it is done distinguish the elite interfaces from the rest.

To see if your interface works properly at higher rates, use a spectral analyzer that shows the whole available spectrum and record your mic at 96. See weird stuff up near 40? 

I have seen several interfaces including Audient that tried a poor filter at 20k when doing 96k recordings. There was tons of noise then bouncing to the max signal level of any frequency near the nyquist. By this i mean, the lines between 20 and 40 were mostly quiet but near 40 there would be as much noise as the noisiest audible frequency.

Sent it back twice and it now works properly. Records dog whistle pieces beautifully.

There are other interfaces identically goofed so it must be a common engineering mistake.

For fun, you can use an eq to remove all content below 20k. Bounce. Then lengthen the recording without snap points for a perfect stretch. This will bring that content to the audible range for your enjoyment.

You can also do the eq after the stretch instead of before. The benefit of before is that you know where the 20k point is. ?

Edited January 21, 2020 by Gswitz

[John]

Depending on how old one is it is highly unlikely one past 20 or so has hearing that does come close to 20 to 20 KHz. Our hearing over time degrades. This is normal. The older you are the less chance you can hear very high frequencies. Most humans cut off around 15K when older. This becomes more pronounced the older we get. If one hasn't been very careful with protecting ones ears  from very loud sound that in and of itself can damage hearing greatly.  Saying that it becomes hard for me to believe anyone over middle age has hearing acute enough to make any definitive statement regarding the merits of high sample rates through listening.

I am 70 and though I hear well for my age I also know people that are younger than me that can't hear music that I have no trouble hearing. Still I would never put my ears up against a 20 year old or younger when it comes to acuity.  

Also it is true we have learned to hear what we remember even though actually we are no longer able to hear what we think we hear.  A simple test  is using flat frequency headphones and use a sine wave generator and sweep it up to where no sound is heard. Check what that cutoff is.  I'll bet most can't get beyond 15K.

BTW it is a very complicated subject and I have only touched on the most basic points here.  There are a slew of variables that impact all sorts of points regarding human hearing.  

[bitflipper]

2 hours ago, Starship Krupa said:

I've not looked into their descriptions, but yeah, it sure would be nice to know whether the record in question was actually mastered at that rate. ?Otherwise, as you say, even if the buyer does think that there's some benefit to be gained by playing a higher bit rate file back through their fancy DAC, they could get the same effect by running their 44.1K or 48K lossless file through a good resampling program.

And when I say "good" resampling program, I only wish that every audio program could up and down sample audio at every rate without bungling it. The fact that not all of them do is an empirically proven fact. Our dear Cakewalk's resampling algorithms handle the job in pristine fashion.

 

True, Cakewalk's resampling algorithm is quite good. However, bear in mind that a "good" resampling algorithm isn't good because it improves the sound quality, but rather because it doesn't degrade it. Best-case scenario is that the upsampled version sounds exactly like the original.

Even if those premium "high resolution" records were mastered at 192 or 96 KHz, it still wouldn't make any difference because what matters is the sample rate they were recorded and mixed at. Once recorded, it's not possible to improve them; all you can do is avoid making them sound worse. Remember, the only thing higher sample rates do for you is extend the frequency range. If there wasn't any >20KHz content in the original recording, upsampling isn't going to magically put some in.

 

[micv]

When I export to 96k, my soundcard got changed to 96K.  How is it possible that the project is running at 48k and the card at 96K?

 

[bitflipper]

That's surprising. The sound card doesn't have to change for an export, since the sound card isn't involved in that process.

I guess it does that in case you want to actually hear your mix after exporting it. Maybe you're doing an audible export, meaning listening to it as it saves the file?

Either way, it shouldn't change your project's SR. It'll still be 48 KHz and when you hit the spacebar to play it back your interface should revert to 48K.

[Starship Krupa]

10 minutes ago, bitflipper said:

True, Cakewalk's resampling algorithm is quite good. However, bear in mind that a "good" resampling algorithm isn't good because it improves the sound quality, but rather because it doesn't degrade it. Best-case scenario is that the upsampled version sounds exactly like the original.

Even if those premium "high resolution" records were mastered at 192 or 96 KHz, it still wouldn't make any difference because what matters is the sample rate they were recorded and mixed at. Once recorded, it's not possible to improve them; all you can do is avoid making them sound worse. Remember, the only thing higher sample rates do for you is extend the frequency range. If there wasn't any >20KHz content in the original recording, upsampling isn't going to magically put some in.

 

As I described, "pristine," meaning clean, without adding or subtracting anything. Put in a 10Hz-20KHz sweep recorded at 44.1K, convert it to 96K, compare the two waveforms using the appropriate analysis tools and they should match.

By "mastered," I meant whether they had had some kind of final mixdown or processing done at that rate, for whatever reason.

On another subject, an aspect of audio perception that I just thought of that I have never seen discussed is possible differences in people's ability to detect transients. I've never been tested for anything like that by an audiologist. They put headphones on me and run a series of pure monophonic sine waves and that's the extent of my hearing test. It's 100% frequency response.

Seeing how our hearing evolved in an environment that was almost entirely devoid of pure sine waves, yet filled with directional transients that were important to hear if one were to survive, maybe there's more to the story than can be learned by playing sine waves to people. Is it possible that frequency response has been deemed the thing that matters because that's the thing the researchers know how to test for? ? It wouldn't be the first time....

[micv]

4 hours ago, micv said:

When I export to 96k, my soundcard got changed to 96K.  How is it possible that the project is running at 48k and the card at 96K?

 

My bad, I must have seen stuff!  I tested it again and the soundcard always locked to the project rate which is 48k even when I do audible bounce.

[azslow3]

Since no one has mentioned that... Audio interfaces can work at 96kHZ or more, so with 48kHz waves, for input and output. But most (exceptions?) drivers physically can not reproduce more the 21kHz. The result is the same as with audio interfaces working at lower rates: analog chain can have LPF so what can not work is cut or follow "lets hope the user knows what he does" scenario. I can not find the links now, but I have seen some tests in the Internet. The same kind as in this thread videos, but checking what really comes out of monitors/headphones when the signal includes over limit frequencies.

[bitflipper]

12 hours ago, Starship Krupa said:

On another subject, an aspect of audio perception that I just thought of that I have never seen discussed is possible differences in people's ability to detect transients. I've never been tested for anything like that by an audiologist. They put headphones on me and run a series of pure monophonic sine waves and that's the extent of my hearing test. It's 100% frequency response.

Seeing how our hearing evolved in an environment that was almost entirely devoid of pure sine waves, yet filled with directional transients that were important to hear if one were to survive, maybe there's more to the story than can be learned by playing sine waves to people. Is it possible that frequency response has been deemed the thing that matters because that's the thing the researchers know how to test for? ? It wouldn't be the first time....

I like your logical thinking. However, there are some fundamental truths of physics that, if taken into account, might nudge your thinking in a slightly different direction.

First of all, let's dismiss the often-heard argument that sine waves don't exist in the real world. They absolutely do, and in fact even the most complex sound can be shown to be constructed entirely of many sine waves. Refer to the groundbreaking book On the Sensations of Tone as a Physiological Basis for the Theory of Music by Hermann von Helmholtz, which you can read for free here. What makes it such a great introduction is that it was written in 1863 when nobody would have had any idea what he was talking about, so the explanations are given without any presumptions about what the reader already knows.

When you see a transient, or any abrupt change in level, think of it as containing high-frequency content. When I was in electronics school, my instructor had us add sine waves by hand on graph paper. It was a tedious exercise but very enlightening. As I kept adding harmonics and plotting the algebraic sum of them, the resulting waveform took on new but familiar shapes. Depending on the harmonic relationships, I got a square wave or a triangle or a sawtooth. I then experienced an epiphany about how subtractive synthesis works: the complex waveforms that we use as raw material for sculpting tones are comprised of many frequencies (sine waves). And that the steepness of the leading edge of a square wave increases as you add more and more high frequency harmonics to it.

Later, I went to work as an instructor at that same school teaching oscillators, amplifiers and filters. Many of the experiments I devised for my students revolved around my personal favorite topic, audio synthesis. We'd run a square wave oscillator into a low-pass filter to show how the leading edge got more rounded as you lowered the cutoff frequency. As well as proving that a truly square shape as often drawn in diagrams can't really exist in nature because it would require an infinite number of harmonics.

And the most important lesson: showing that until you rolled off at a point below the upper limit of hearing, there was no audible difference in how the "square" wave sounded. Removing, say, 30 KHz from the signal made no difference in how the sound was perceived. However, the effect was clearly visible on an oscilloscope. Removing frequencies above the hearing range obviously changed the waveform, but did not change how it sounded.

This is why we can safely ignore frequencies above 20 KHz in digital audio. That's fortuitous because the sampling theorem only applies to a band-limited system. If it was necessary to preserve ultrasonic content, you'd need a much, much higher sample rate. Digitize a 20 KHz square wave and you get a 20 KHz sine wave. But both sound exactly the same (assuming you can hear them at all). Digitize a 12 KHz sawtooth and you get a 12 KHz sine wave - and they both sound the same because the sinusoidal components that distinguish a sawtooth from a sine are above the range of human hearing.

This is all a long-winded explanation for why a hearing test using only sine waves is valid.
 

 

[Starship Krupa]

I get you, and thank you for reminding me of Heimholz and On the Sensations of Tone and how any waveform can be created using sines. In my musings I had sort of forgotten about ol' Hermann's work.

I would like to point out that I did not say that "sine waves do not exist in the real world," I said "almost entirely devoid of pure sine waves," which is different.

I also did not mean to imply that the standard hearing test was somehow not valid nor suited to its purpose of measuring hearing loss, although I can see where it looked as if I was.

What I was contending, and still do, is that when discussing audio reproduction and perception, there may be abilities or acuities we haven't studied on the human side, and things on the reproduction side that relate to that.

We already know, for instance, that different people have different curves and different ranges. What if we also have differing degrees of sensitivity to IM distortion or group delay or phase coherence or transient sharpness or whatever? Maybe researchers are studying that, I don't know.

The most prominent blind test I've seen for whether people could choose lossless vs. lossy music files, the NPR one, was completely flawed due to the delivery system being a web browser. Yet hundreds of thousands of people accept it as proof that even recording engineers can't spot the difference. Well, not through Firefox I can't.

I am a skeptic. I accept 100% that recording at 96K should make no difference whatsoever in what a normal human can hear. My hearing rolls off around 12K anyway. I record at 44.1. I am also open to the possibility that recording at higher rates may have some side effect that makes the audio sound better to some people. Maybe not all people, maybe just people with the ears that are extra sensitive to whatever.

50 years ago guitar amplifier designers and musicians and audio engineers were told that vacuum tubes would soon be phased out in favor of solid state devices, that transistor amplification was in every way superior, if they heard any difference it was imaginary, or the solid state device would sound superior, etc. and so on. The only people who bucked this were the musicians, and since they were all on drugs anyway, nobody paid them any mind.

I will not bore you with recapping the story of how that turned out, but I will say that I have made a good living recapping tube amplifiers from 50 years ago.

[Gswitz]

15 hours ago, azslow3 said:

Since no one has mentioned that... Audio interfaces can work at 96kHZ or more, so with 48kHz waves, for input and output. But most (exceptions?) drivers physically can not reproduce more the 21kHz. The result is the same as with audio interfaces working at lower rates: analog chain can have LPF so what can not work is cut or follow "lets hope the user knows what he does" scenario. I can not find the links now, but I have seen some tests in the Internet. The same kind as in this thread videos, but checking what really comes out of monitors/headphones when the signal includes over limit frequencies.

Hey Azslow!

With lasting respect for you, I took a minute with my RME UCX to do the following to show that RME exports the higher frequencies to the speakers...

1. Dragged a loop into a new project at 96k (shl_guit_80_ringshft_D_Rex(9)) -- noted audio ranging up near 20k.

2. bounced to track to get an audio file I could use the loop construction view on.

3. Used Craig's Les Paul trick from p270 from the Big Book of Sonar Tips to speed it up 7 half steps.

4. sent the stereo pair out of the interface and back in physically.

5. Recorded the raised recording and noted content up to around 40k.

6. Next, I slowed down the new recording by 7 half steps and did a null test with the original. It was pretty good to around -33dB.

So this shows that through this process, most of the audio was retained. This defends that the RME UCX will send the higher frequency content to your speakers.

If you were questioning whether my speakers do a good job up there... haha... not worth testing. ? I have cheapie charlies... but I'm sure you can find speakers that attempt it anyway.

As an interesting aside, while the first and final spectral analysis bits are almost identical, at one moment you see a tiny jump at 40 (otherwise quiet between 20k and 40k like the original). So, not perfect null test as I said. Something funny in there.

No other gear or interface in the middle. Out of the RME UCX on channels 1 and 2, in on channels 5 and 6. No pres on those channels.

Because that final 40 was after slowing it back down, I suspect something to do with the loop construction view slowing algorithm. Who knows what. ?

Edited January 23, 2020 by Gswitz