Been bitten by the 96 kHz bug...

[mikehalloran]

I have been interested in this subject since I was a kid. My dad who worked at AMPEX had a series of Full Frequency early stereo recordings - I still have some of them.

I remember one, The Best of The Dukes of Dixieland, a live album which had a technical note on the back stating that, 'under a microscope, one could see frequencies down to 16 and up to 25Kcps (no one called it Herz back then). Even though the ear couldn't hear those extremes, they felt it was important to leave them on the record'. OK... that got my attention and I never forgot it. A set of headphones puts you in the audience - it's still a great album.

Years later, studies were done to show that, even though we can't hear those high freqs, the brain does use them to locate stereo images. This is one of the reasons that true stereo sounds better in good analog than CD where frequencies above 19K are subject to a non-linear roll off to prevent certain artifacts.

Another issue is the math involving sampling of high frequency harmonics. A 15K cymbal harmonic has twice as much information at 96K and 4x as much at 192K.

A 96K stereo sample sounds better than 48K and 192K sounds even better.

None of this explains to me why a mono or stereo 48K sample would sound better at 96K. I am not certain that having twice as much high harmonic info artificially generated would do it - but perhaps it does.

If you ever get to hear the Mobile Fidelity Labs vinyl recording of Abbey Road, the first thing you notice on Something is that no CD in your collection - including the remastered Abbey Road - has cymbals that sound as real. Why there is no SACD version when hi-def analog reveals the possibilities is a mystery to me.

[MIDI Life Crisis]

I hear the MP3 difference from the WAV files but frankly don't hear enough of a difference between the two WAV files (and for that matter the two MP3 files) to make me what to record in 96k v 48k.

Thanks for posting those.

[Kubi]

None of this explains to me why a mono or stereo 48K sample would sound better at 96K. I am not certain that having twice as much high harmonic info artificially generated would do it - but perhaps it does.

My guess is it's bc the interaction between so many harmonically rich 44.1 kHz recordings at various levels ends up being quite complex. Also not sure what kind of internal processing kicks in inside Kontakt 5 at higher SRs. At least that's my explanation. I purposely chose to test this with samples, since it's obvious that live recordings would benefit and since I do use a lot of samples.

I'm sure in a full mix the differences only get more pronounced - the thing that kicked this whole test off was a comment by Gregory Scott of Kush Audio, who said the highs on his new Clariphonic plug-in sound especially good at high SRs. So apparently, even if the source samples are lower res, processing them improves audibly at higher resolutions. Similar things could already be happening inside Kontakt 5, on playback at various levels, and as they get densely stacked using pedal.

Ultimately I don't know what's happening exactly, but the ears make the final judgement. I think the diffs are quite pronounced.

[MIDI Life Crisis]

...the ears make the final judgement. I think the diffs are quite pronounced.

Could be. LOL! Thanks again.

"... I beg of you to keep the matter of my deafness a profound secret to be confided to nobody, no matter whom..." ~ Ludwig von Beethoven

I guess that's why he decided to go into manufacturing drums.

[mikehalloran]

96K certainly doubles the freq at which the shelving filter kicks in and takes its effects to the inaudible range. Could it be that simple? Perhaps.

[Phil O]

So I downloaded the two wav files and have listened to them. Certainly different to my ears, although I'm not ready to say I like one better than the other - just different.

I upped the 48K file to 96K and tried to null the two files in a 96K DP project. No dice. These two files are distinctly different. I can't even find a place where the waveforms appear to line up.

Umm, I've got nothin.

Phil

[kgdrum]

When I worked in the HiFi business we had access to some of the best digital playback equipment in the world.
If you ever hear digital played back on a DCS Scarlatti system or an equivalent piece of gear from Accuphase or Esoteric at 24/96,24/192 or DSD you will be amazed at the nuance,detail,dynamic range and the lack of digital harshness we're all used to hearing.
The reality is most people never will hear music this way or have equipment that can actually play back digital this way or have a system that can actually reproduce the range of fidelity a 1st class system is capable of playing,these digital playback systems can approach $60,000 or more!
You need an entire system capable of playing the material full range when the source is high resolution.
I have by HiFi standards a modest system,my DAC can play up to 24/192 files natively and up-sample redbook files up to 24/192
It's very good & it's hooked into a very nice system well beyond most peoples HiFi systems.
If you were to playback 16/44 files or 24/44 files and then playback native 24/96 or 24/192 files or even up-sample the red book files to 24/192 as I do all of the time,if you don't hear the differences I'd suggest you get your ears checked!
For me I hear more dynamic range,less harshness from the filters,better imaging and more of an "analog type sound".
I don't think we actually hear the super higher frequencies but we do sense the nicer more open imaged sound,better dynamic range and lack of hash associated with bad digital.

[Kubi]

Except the difference is still audible in the MP3s, albeit less so. Which to me means it's worth doing during production even for real-world applications. Much like we're using 24bit WAVs for production even though most end users will listen as 16bit MP3s.

[TnMike]

I want to try this with my Ivory piano. So I record a MIDI track of piano, freeze it with the sample rate at 48k and the bit rate at 96k (using my Apogee external clock).
Then play back at 44.1/16.
Do I have this right?
Thanks,
Mike


G5 2.7 GHZ, DP 7.24, MOTU 2408, Ivory, SD2, B4ll, UAD-1, Melodyne, Ramza DA-7,

[Kubi]

Not quite - you have to set DP to a sampling rate of 96 kHz (can be clocked internally if you like, or externally, just set DP accordingly) then record the track (label this Ivory 96k).

Then you set DP to a sampling rate of 48kHz, record another track (label this one Ivory 48k).

Easiest place to set sampling rate in DP is on the right-hand side of the main transport controls.

You end up with two different audio files, one at 48kHz and one at 96 kHz, each labeled accordingly.

[TnMike]

Yes, I'm sorry...I meant freeze track at 96k. Then freeze another track at 48k. Then playback both at 44.1 or 48k...
and compare?


G5 2.7 GHZ, DP 7.24, MOTU 2408, Ivory, SD2, B4ll, UAD-1, Melodyne, Ramza DA-7,

[Phil O]

This is what I don't understand at all.... If you say the instrument was originally recorded at 44.1 KHz but sounds better listening to it at 96KHz, where does the "improved information" come from??
How does it improve the signal if the information was never there to begin with?

This is really the question, isn't it? As I posted earlier, the two wav files are clearly different. In order to compare them, I had to up-sample the 48K file to 96K (up-sample - is that the correct term?). I don't think the sample rate conversion should have changed the file that much. So where does the difference come from?

The only two things I can think of is 1) the VI software is handling something differently at different sample rates or 2) DP's summing is significantly different at different sample rates. If it's 1, then different VIs may yield different results. But, if it's 2, that's something for all DP users to consider. One way to check it would be to start with a project of just audio files recorded at 48K. Mix it down at 48K, convert them all to 96K, mix it down again, then compare the two. Any takers?

Phil

[Kubi]

If it was a single stereo 48kHz file at unity gain, there would have been no measurable difference. (Other than perhaps a few artifacts from the SR conversion that would be inaudible.)

But a Kontakt patch playing back a polyphonic track of a piano-like performance of tons of notes with pedal is a very dynamic mix. Now all these 48kHz samples can start and overlap at various points in time on a much finer grid than a 48kHz environment would provide. Their interactions in the summing also now don't have to be chopped off at 24 kHz or thereabouts, they can be left untouched way into the supersonic range. It's of course also possible that Kontakt itself indeed behaves differently in higher SRs, i.e. maybe there's a Nyquist filter inside Kontakt as well? Wouldn't know. Regardless, it's the fact that playing back a sampled performance is the equivalent of live-mixing a dense track - much more complex than the original resolution of its parts would allow. Thats why I think it makes a difference even thought the source samples are 44.1 kHz and there is no additional processing.

Now, I'd put money on the theory that a 48 kHz mix remixed at 96kHz sounds better than a mix done in 48kHz all the way. Why? Because at least some (if not all) plug-ins behave differently at higher SRs, taking advantage of the extended Freq response that's possible. In fact, my whole 96kHz test was a reaction to a comment made by Greg Scott of Kush Audio re. how his upcoming Clariphonic plug-in sounds even better at high SRs because the highs can be smoother. Also of course, the interaction between the tracks in DP's summing can be executed at higher resolution. Bound to sound better. I'll have one that needs doing sometime soon (a 44.1 kHz project already tracked and ready to be mixed.) I may print it at two different resolutions just to test.

[Kubi]

Yes, I'm sorry...I meant freeze track at 96k. Then freeze another track at 48k. Then playback both at 44.1 or 48k...
and compare?


G5 2.7 GHZ, DP 7.24, MOTU 2408, Ivory, SD2, B4ll, UAD-1, Melodyne, Ramza DA-7,

Easiest is to play them back in Quicktime 7, then you can have both open at the same time. In DP, you'd have to run a SR conversion on whichever track doesn't match the Projects SR first, which will change the sound (and DP's SR conversion is workable, but it isn't as good as i.e. Peak's).