Question about 44.1 sample libraries and 48 KHz DP sessions

[Shooshie]

I hadn't thought about this. I did some stuff at 48 a while ago and thought it sounded better. I've been changing the templates and settings to 48 and hadn't considered rate conversion overhead. And now I think about it, tracks from VIs recorded at 44.1 can't sound better -- same at best and possibly worse.

[Edited]
It would make sense to put out a library with higher sample rates, down sampled to 44.1K, rather than the opposite.

I seem to remember some VI's having sample rate choices. I've always wondered if they actually had samples at those rates, or if they are doing a conversion. I had assumed they start with higher rates and down sampled if needed, but I don't know if they do. I'm researching it right now; I'll let you know if I find anything about it.

Shooshie

[FMiguelez]

It would make sense to put out a library with higher sample rates, down sampled to 44.1K, rather than the opposite.

If I understand correctly, that's exactly what VSL did... They recorded at 24-96 but deliver at 24-44.1

[Shooshie]

It would make sense to put out a library with higher sample rates, down sampled to 44.1K, rather than the opposite.

If I understand correctly, that's exactly what VSL did... They recorded at 24-96 but deliver at 24-44.1

But the question is: what are the sample rates for VSL on your hard drive? Do they ONLY deliver 44.1? Or do they deliver a couple choices, switched when you switch your project sample rate?

With a library the size of VSL, it would seem impractical to deliver anything BUT 44.1. If it's recorded at 96K, then whatever it benefits it provides for the sound in the audible range will still be captured if they convert to 44.1K, minus the high resolution of frequencies around the threshold of hearing. So, if you hear something really good in the audio recorded at 96K, whatever you are hearing has trickled down to the audible range, and that range can be captured at 44.1K.

Now, using it at 44.1K, all the frequencies and harmonic interactions and modulations at the Nyquist frequencies will produce aliasing. If you raise the sample rate to 48K, you can lift the Nyquist frequencies well above the hearing range, giving some definition to all the audio that is in the hearing range, or close to it. That gives the audio some room to interact and create combination and difference tones, the stuff that fills up musical harmony with content that's not "really" there, but which adds considerably to the harmonic content and overall pleasing sound. Without those upper partials, there would be less of that. How much less, and how more or less pleasing the sound might be, would be a matter of perception and opinion, so there's probably no way to get a solid, scientific answer out of this.

In the time it takes to read this entire thread, you could probably record some things at different sample rates, and figure out whether you can hear any differences. I'd play it for some young people whose hearing is still intact, and see what THEY say. In the past 5 years (ages 55 to 60), I've lost almost 4000 hz of hearing range. Some of that comes and goes, but for dependable hearing, I can only count on about 14,000 hz. Sucks, but that's one of the lesser challenges of aging!

Shooshie

[mhschmieder]

I understand what the article's saying, but they missed some important points, several of which argue in favour of 88.2 or 96.

Nowhere does the article mention virtual analog synths, where indeed the best sound by far is generally at 96 kHz (maybe also 88.2 kHz but I never bother with that sampling rate).

I consistently get better results every time, using such sound sources.

For sample libraries such as VSL, I have done many many A/B's over the years to put my mind at ease, and rarely detect any difference even though on-the-fly conversion may be taking place.

If recording using mics, or even direct line for analog synths or organ etc., I notice a BIG difference at 48 kHz, for most sound sources. I have always assumed it MOSTLY has to do with anti-aliasing of the filters in the higher range, and that the extra headroom of 48 kHz pushes the anti-aliasing further beyond the range that we either hear directly or perceive as air.

OTOH, many years ago, when I first started with digital recording (using the Alesis MasterLink for several years, as I didn't own a computer for a LONG time), I detected VERY strong differences on cymbals, acoustic guitar, and anything else having complex high frequency sustain, if using 96 kHz.

This brings us to an important point that is so often neglected in these articles: it isn't just about the oversampling rate, but the actual SPACING of samples (in the Time Domain), and thus how smooth and detailed the discretization is of what would otherwise be a continuous signal.

Granted, the majority of material and recording techniques do not really "need" 96 kHz and get most of their benefit from 48 kHz. So I really take it on a project by project basis.

My own rule of thumb is to use 48 kHz for anything specifically targeted towards video, and then either 48 kHz or 96 kHz for my own projects depending on instrumentation and the ratio of Real Analog synths to VI's to sample libraries to miked instruments.

I use 44.1 kHz when someone I am working with won't budge on the matter. I never get into an ugly fight over it; I simply raise the simple arguments (similar to those in the linked article) in favour of 48 kHz and usually win.

Luckily, none of the projects (several of which are my most active ones currently) where I am at 44.1 kHz suffer dramatically for not being at 48 kHz. It's not a huge deal for me; I just like to have a consistent rationale so that I remember the reasons why I pick one over the other.

Interestingly, the only lingering question that I had not felt was completely resolved was the very one brought up here, regarding the native format of various sample libraries.

As there is no consistency and as I use so many (even within the same project), I have never felt I could standardize on that principle alone, so I have given up on having it feed into the criteria -- especially as I have not personally experienced degradation from on-the-fly conversion of the samples (and this is coming from someone who was gifted with two sets of genes that resulted in abnormally good ears).

It seems like we have come to a consensus as a forum on this topic, more or less. That is, that it's usually best to use 48 kHz if practical, and to not get hung up otherwise.

[monkey man]

... discretization...

Word of the year thus far for me, Mark.

[mhschmieder]

There were a lot of things covered in this discussion, and when I woke up this morning I realized that I had missed an important point regarding using Vienna Ensemble to host and route a whole bunch of channels simultaneously.

I render as I go, for a variety of reasons, but may eventually adopt the VE approach, so now that gives me pause. When I said I don't notice any major differences with VSL in a 44.1 vs. a 48 project, I meant in the context of how I work, which is to render each track individually and during development of the project.

My understanding is that most people only do one final gigantic rendering when finalizing a project, if even then. I appreciate the reasons many people work this way, but for me there is one reason more than any other that trumps everything, which is the need for future-proof projects where every track has a rendering in a standard audio format that can be used almost anywhere even if the host program or the VI's etc. go away over the long haul.

Anyway, given that many people do leave all their tracks as live MIDI and render them all at once via massive buss routing or VE instances, the likelihood that many simultaneous real-time sample rate conversions could degrade audio quality is quite serious indeed.

I look forward to hearing what people discover about this situation as they experiment and narrow down where the break off point is in terms of simultaneous track count for real-time sample rate conversions of sample libraries.

[billf]

It would make sense to put out a library with higher sample rates, down sampled to 44.1K, rather than the opposite.

If I understand correctly, that's exactly what VSL did... They recorded at 24-96 but deliver at 24-44.1

Here is an example of a vocal sample set I looked at recently. 48k seems to be a growing trend for new sample products.

Product Specs

6.32 GB
8,310 Samples
38 instrument presets for Kontakt 5.1
24 bit / 48kHz uncompressed PCM wav samples
Unlocked Kontakt presets and samples that you can customize

And they record at 48k now:

Most of our products include standard pcm wav format audio files. Most of our products are recorded in stereo at 24 bit / 48kHz, although some of our original libraries were captured at 16 bit and/or 44.1 kHz. Nearly every modern professional audio program in the world can load wav files, so even if you're not a Kontakt user, you can still enjoy our outstanding instrument and vocal sample content in an environment that works for you.

[FMiguelez]

The more I read about this, the more I realize it's mostly crap that banks on our subjectivity and gullibility. I mean, if you consider there are people in this world who claim to communicate with the dead, and hear voices in their heads, then it's no wonder we think we hear diferences where there may be none.

Consider that the food you ate 1 hour ago can have an impact on your hearing, and if you're stressed or moody as well.
A simple and subtle EQ setting will have more impact (good or detrimental) on the sound than recording a hi-hat at 96 KHz or 44.1.

I did a few A/B tests, and my results showed no difference between a random answer and a consistent one.

I think we all are biased and are cherry pickers and fall for confirmation bias fallacies more often than we don't...

[mhschmieder]

That may be true for some people and in some situations (especially if someone just dropped a load of cash on something), but in my particular case, the confirmation bias was the opposite of the results.

I haven't run those specific tests since getting a computer though, so I may have been doing nothing more than finding out how each sample rate performed specifically on the Alesis MasterLink, which of course is so many years ago as to probably be laughable in terms of relevance to current debates.

Like I said though, I don't bother even if I think it's better to do 96 kHz, unless running a solo project that makes heavy use of virtual synths where the algorithms are really dependent on the high sample rate. This was even a requirement for Sample Modeling's saxes up until they switched to SWAM.

The latency at 96 kHz is an advantage, of course, when overdubbing extremely time-sensitive stuff. Having said that though, my own company has the lowest latency in the industry for its own DSP stuff, even though we were only running at 48 kHz. So indeed, latency is also a factor of algorithms, technology, and the chips involved. It's good to know what the ultimate limiting factor is, and that continues to improve over time.

I personally think the trend will be more towards 48 kHz, with 96 kHz as an option, when it comes to content such as sample libraries. Just as we would hear arguments fifteen or twenty years ago about why it is more practical to stay at 16 bits, but eventually everything but CD's and MP3's caught up to the 24-bit trend, and even a lot of sample libraries are going that way.

There's no one choice that can solve every need or every situation, and that's likely to continue to be the case. For each project, I think we just have to be aware up front what the likely targets are, the contributors and what limitations that places on collaboration, etc.

Other than for 16-bit recording, I just don't think sample rate is the biggest issue on most projects. I wish it were possible to consolidate to one standard, but we've already seen how many factors are involved, and that we also can't afford to get overly worried about real-time conversion when recording sample libraries either as there is rarely a project that will luck out and be using just one source.

I would think that the disc-streaming algorithms and other stuff would be more intense than the conversions anyway, so other than for huge VE Pro templates all triggering MIDI together in real time on each playback (i.e. no frozen or rendered audio tracks), I doubt it's worth getting worked up about or changing project policy for.

[monkey man]

Mark's modesty in this context is impressive; I've half expected him to mention this but he's been able to resist:

He's always been able to hear significantly higher frequencies than most of us (at least to 22kHz). To someone like him, the 44/48 difference would surely be obvious; even the 48/96 one, although less so I'm guessing, would nevertheless be readily apparent.

Just sayin', 'Nandito.

On another note, Mark:
I'd have thought that whether one rendered all those VI tracks separately or as a mix, the resulting 1s and 0s would to be identical - a matter of CPU cycles and nothing more.

You indicated that errors are more likely in the latter scenario, and I'd love to be enlightened as to how this could be the case. It's been my understanding that if an app such as DP isn't able to accurately render something, it must've hit either a disk or CPU limit, in which case it'd warn us.

Don't get me wrong mate (a caveat I'd not normally add, admittedly, but the "heat" in the forum of late has me tip-toeing on eggshells), you know I respect you enormously (probably the wrong word, but you know what I mean), and this is why I'm asking. I mean, if anyone can explain this to me, you'd be one of the few I've met who could.

Thanks mate.

EDIT: I misplaced the quote and replied from memory, so I think I've missed the point somewhat. Here 'tis:

Anyway, given that many people do leave all their tracks as live MIDI and render them all at once via massive buss routing or VE instances, the likelihood that many simultaneous real-time sample rate conversions could degrade audio quality is quite serious indeed.

I think what you might be saying, Mark, is that the compounded SR-conversion distortion could be an issue, at least, when compared to your method.

My question still applies 'though: How does rendering the tracks and therefore SR-converting them separately, followed by mix summing, differ from processing the whole shebang at once?

[FMiguelez]

I know what you guys are saying.

BTW, I didn´t mean to sound like an arse... I think my last comment came out a little... ... stronger than I meant. Sorry about that (the about meant with a Canadian accent).

I don´t doubt there are people who can hear all the way up there.

What I really wanted to say is that, there are still a million things one could do to ruin an otherwise "perfect" mix that would have much more impact than the sound of the sampling rate alone.
These things should be taken care of before worrying about the subjective differences that different sampling rates MAY have.

The way the brain translates and interprets the sound waves the ear catches could be the very definition of subjectivity. ¿Remember the blue dress?

Also, worrying too much about these little things can be a waste of time and energy. That effort is better put to make actual music

I´m just glad I think I made an informed decision, and I´m happy I can finally archive this in the back of my mind and not even think about this anymore (until some future technology makes an obvious improvement I should look into, of course).

Yup. I noticed the forum has been a bit hot as of late, Nicky... Bananas might work better for you than eggshells when walking in the MOTUNation heat, though,

[MIDI Life Crisis]

It's a bit (pun intended) like the audiophile debates. In context, probably 99.99999999% of produced "music" is NOT listened to in a vacuum or even desirable conditions. After factoring in the defects inherent in the final playback system, the context (surrounded in a bed of dialogue, SFX, bgr in music for picture - for example) and actual listening environment noise and room characteristics, not to mention noise generated by and for, and audio deficiencies of, most listeners, it doesn't make a rat's ass bit of difference if you're at 44 or 48 and whether some guy in a perfectly tuned studio can hear the ping sustain in a splash cymbal at or near the limits of human hearing.

Where it makes a difference is how the technology integrates and interpolates the physical ANALOG, essentially mono, sound source that actually exists in nature and turns that into something acceptable for binary creatures, most with two ears and with each ear needing to hear things a little different (lest we get consumed by a saber tooth transient).

Point being, yes: be anal retentive in producing the best sound you can, but understand it's not ultimately how "clean" or "perfect" that sound is but rather what the underlying expression in the "art" communicates to the neanderthal snorting and sneezing while your splash cymbal decays.

It's all relative.

[doodles]

My proper audio friends always call me an ass, as they work at 48k in recording studios!

But I run everything at 44.1 (including both slave machines running VE Pro), as I incorporate lots of old audio from previous sessions, loops, samples, etc. i just export at 48k, but do everything in the box at 44.1k/24 bit. For me the essential part is the 24 bit. Certainly for soundtracks. 16 bit just so top heavy.

[Shooshie]

It's a bit (pun intended) like the audiophile debates. In context, probably 99.99999999% of produced "music" is NOT listened to in a vacuum or even desirable conditions. After factoring in the defects inherent in the final playback system, the context (surrounded in a bed of dialogue, SFX, bgr in music for picture - for example) and actual listening environment noise and room characteristics, not to mention noise generated by and for, and audio deficiencies of, most listeners, it doesn't make a rat's ass bit of difference if you're at 44 or 48 and whether some guy in a perfectly tuned studio can hear the ping sustain in a splash cymbal at or near the limits of human hearing.

Oh, man. What a relief. I can stop worrying now!

Nevertheless, I just prefer 48K/24bit. Sounds more restful to my ears for some reason. Once in a while I accidentally get set to 44.1K, and pretty soon, without really knowing why, I'll find myself looking for why my audio sounding just a little on the harsh side. I always find it, and when I do, I switch it back to 48K, if it's possible at that point, which is a pain in the assets, and sometimes requires re-recording some stuff. Logically, it should make no difference, and I should not be able to hear that. But apparently I do, and it always feels like "ahhh" when I finally get things back to 48K.

While listening to CDs at 16bit / 44.1K doesn't bother me on stereos, in cars, and such, if I listen to them on my computer, they bother me. I long for a turntable, which I'll get when I finally move to the ranch, but I don't really have room for one in here. I just wish they'd record everything again at 48K, and make that the new standard for CDs. But then someone would say "why not 96K?" Well, it has to stop somewhere, but why NOT make it 96K? Just remove the whole debate.

Short answer, "because that ship sailed in about 1980." I made my first digital recording with the Dallas Symphony (as a musician, not an engineer) in 1981. At that point, it was already too late to change the format. The standard simply came before the hardware was ready. The digital recorders they were using for the Dallas Symphony at that time were in racks about the size of a washing machine. We had to wait a while when they were probably printing the tracks to a hard drive, which was probably a huge Winchester drive or something. If they'd only revised the standard 10 years later.

Shoosh

[MIDI Life Crisis]

As mentioned earlier, I'm always at 48/24 on my projects. I can tell when an app plays at 44.1 and it doesn't bother me, per de, but for any pro applications - especially working to picture as I do - it's a no brainier. Set it at 48 and forget it.