The 96Khz Question

[Bela D Media]

This is a mono source instrument. It will be ram friendlier then stereo but will use an IRF. 2-4 Gigs slated.

?/24-bit

Thanks,
FB

[Bela D Media]

everybody I run into these days seems pretty happy listening to data scrunched AAC and .mp3 files on their iPods. Hahahahaha!! :D ...and I agree that 48khz seems to be middle ground.

[Nick Batzdorf]

A friend of mine puts it this way: "96k can kiss my axx."

:)


Recording at 192, sure. But sample libraries should wait until RAM access and processing power are no longer such precious resources.

[Veron]

I thought 44.1 is enough.
One reason is because of the RAM and CPU usage.
Or maybe 48 might be fine if it doesn`t effect the processor usage that much.
The best thing is, you can record and keep the source file in 96,
and distribute it in 44.1 or 48 at this time, and when computers are
more powerful, you could make a reasonable upgrade path to 96
for the people who want it :)

[faintwhitenoise]

I would throw out the point that resampling from 96k+ down to 44.1k will sound worse than if you'd recorded in 44.1 throughout. That being the case, if your end result is 44.1, then why record at 96k? Now if your end result will be DVD-A or some other high resolution medium, and you have the technical facilities for it, then why not?

[mathis]

I would throw out the point that resampling from 96k+ down to 44.1k will sound worse than if you'd recorded in 44.1 throughout.

There are varying opinions on that. I'd say record in high resolution, but distribute for now with standard 44k1.

[JonFairhurst]

Most film-work is done at 48, and CDs are 44.1; however, many people just run their samplers at 44.1 and fuggetaboudit.

If this is along the lines of bass/drums/guitar, go for 44.1.

If it's a vocalist or some other instrument that you expect to be mainly for film and TV, go for 48 if you're a purist, and 44.1 if you're a pragmatist.

Many systems let you record at 88.2. Maybe some do 176.4? Every time you double the sample rate, you get another effective bit of resolution. That's how 1-bit converters and SACD work. So 88.2 at 24 bits is like 44.1 at 25-bits. Cool, huh?

If this is a sound effects lib, go for the highest possible sample rate. Why? Maybe you've recorded an arrow going "whoosh". The sound designer wants to make it sound deeper and have a longer duration. So he slows it down by a factor of two. One problem... If recorded at 44.1 the whoosh now cuts off at 10 kHz. Had it been recorded at true 88.2, you still get a 20 kHz W-H-O-O-S-H.

But for musical instruments? Go for 44.1 as the default, and 48 if you're targeting film purists.

All the best...

[MMcD]

Even if you recorded at 44.1, you won't have a direct 44.1 throughput if you're using a sampler. Samplers change sample playback frequency to interpolate notes and deal with tuning.
While that is true with hardware samplers with analog outputs, in the case of software sampling, anytime there is a pitch change the source wav is resampled. In most samplers the quality of this resampling is pretty lousy.


I've never done any serious testing, but I've always had the feeling that samplers trash audio (to a greater or lesser extent). Some playback plug-ins, like Intakt, severely mulch audio. Running audio through any sampler software sucks some of the life out of it.
Agreed - if there is any pitch shift for tuning or transposition the quality is reduced unless the interpolation function uses very high resolution.
SFZ is very good at the highest quality for example.

[Anton Bruckner]

96Khz looks good but I think 44.1 is really enough.

[howardv]

I would throw out the point that resampling from 96k+ down to 44.1k will sound worse than if you'd recorded in 44.1 throughout. That being the case, if your end result is 44.1, then why record at 96k? Now if your end result will be DVD-A or some other high resolution medium, and you have the technical facilities for it, then why not?My experience downsampling material recorded high res is quite the oposite. Particularly if it's recorded and mixed at high res and downsampled/dithered as the last step. But even downsampling on a per-sample basis, there's always the advantage of cleaner filtering in the physics of high-res converters (less phase shift in the audible passband).

As far as destination goes, that's the rub for a sample library. The vendor doesn't know where I'm going to use their samples. Heck, I don't know myself until a job comes up. I would think a vendor would want to provide options as broad as possible to maximize its market.

But I have to agree that the technology might not be soup yet. But its close. Right now my only 96k gs3 piano library pushes my system to its limits. My workaround is to freeze and upsample all my vi tracks to incorporate them into hi-res mixes. But the hassles involved with constantly switching sample rates is a pain in the neck. Right now gs3 will run at high res, even with 44.1k libraries, but the performance hit is substantial. I haven't had much luck getting Ni 44.1k libraries to run high res. My hope is that as Win64 and Vista64 establish themselves we might see some higher res sampler and library options open up.

If I was doing a new sample library myself, I'd do as mathis suggests. Sample at 192/24, perhaps with one of those way-cool Tascam, Fostex, or Sound Devices flash recorders. But only make downsampled versions available at first. That would leave high res options open for down the road. Another possibility is the option of recording notes and pharases to analog tape.

Howard

[howardv]

Also, since the bulk of my work is targeted to audio CD, I tend to do most of my recording at 88.2k.

Howard

[Bela D Media]

Most film-work is done at 48, and CDs are 44.1; however, many people just run their samplers at 44.1 and fuggetaboudit.

If this is along the lines of bass/drums/guitar, go for 44.1.

If it's a vocalist or some other instrument that you expect to be mainly for film and TV, go for 48 if you're a purist, and 44.1 if you're a pragmatist.

Many systems let you record at 88.2. Maybe some do 176.4? Every time you double the sample rate, you get another effective bit of resolution. That's how 1-bit converters and SACD work. So 88.2 at 24 bits is like 44.1 at 25-bits. Cool, huh?

If this is a sound effects lib, go for the highest possible sample rate. Why? Maybe you've recorded an arrow going "whoosh". The sound designer wants to make it sound deeper and have a longer duration. So he slows it down by a factor of two. One problem... If recorded at 44.1 the whoosh now cuts off at 10 kHz. Had it been recorded at true 88.2, you still get a 20 kHz W-H-O-O-S-H.

But for musical instruments? Go for 44.1 as the default, and 48 if you're targeting film purists.

All the best...

Dear Jon,

Your post was the deciding factor for me and I thank you for your knowledge. This product is best suited with 44.1/24.

Thank you to all who have chimed in on the subject.

Best,
Francis Belardino
Bela D Media

[Nick Batzdorf]

The reason is the same one that caused them to release the first version in 16 bits. I don't think a statement is necessary.

[JonFairhurst]

Dear Jon,

Your post was the deciding factor for me and I thank you for your knowledge. This product is best suited with 44.1/24.I'm glad to hear that my insights were helpful. (I guess we can infer that this particular product is targeted at the CD crowd, or at least not at film/tv purists.)

Regardless of the sample rate, I expect that it will be a top notch instrument that offers yet another unique sound to the community.

Best of luck with the development of the library.

[Laurent]

I would think that 192 - 96 - 48 khz are going to be the most enduring sample rates as time goes on.

Then again, VSL has got to be approaching a Tb worth of samples, and all their stuff is at 44.1k/24b. I wonder if they record at 88.2 or 176.4. Has VSL ever made a statement on why they chose 44.1?
Do you think "not round" samplerate conversion is still an issue ?
I've heard that know interpolation algorithms are efficient enough to perform as well on 96k=>44.1k than on 88.2k=>44.1k...
88.2k recording would then be an oversampler "early adopter" habit.
What do you think ?

[Richard Berg]

Do you think "not round" samplerate conversion is still an issue ?
Possibly -- no way of knowing what various programs do internally -- but there's no excuse. CPUs are very fast and there are many well-known algorithms for resampling with high order filters.

[howardv]

Do you think "not round" samplerate conversion is still an issue ?Yes. Even-rate conversion minimizes both error and latency. Which may or may not be of overriding importance depending on what you're doing. If you're just monitoring during composing or rough mixdowns, that's one thing. But if you're talking about burning that master that's going into production for all eternity, you might want to take every step, no matter how tedious, demanding, or time consuming, that'll make it the best that it can be.

Howard

[Patthoven]

Most film-work is done at 48, and CDs are 44.1; however, many people just run their samplers at 44.1 and fuggetaboudit.

If this is along the lines of bass/drums/guitar, go for 44.1.

If it's a vocalist or some other instrument that you expect to be mainly for film and TV, go for 48 if you're a purist, and 44.1 if you're a pragmatist.

Many systems let you record at 88.2. Maybe some do 176.4? Every time you double the sample rate, you get another effective bit of resolution. That's how 1-bit converters and SACD work. So 88.2 at 24 bits is like 44.1 at 25-bits. Cool, huh?

If this is a sound effects lib, go for the highest possible sample rate. Why? Maybe you've recorded an arrow going "whoosh". The sound designer wants to make it sound deeper and have a longer duration. So he slows it down by a factor of two. One problem... If recorded at 44.1 the whoosh now cuts off at 10 kHz. Had it been recorded at true 88.2, you still get a 20 kHz W-H-O-O-S-H.

But for musical instruments? Go for 44.1 as the default, and 48 if you're targeting film purists.

All the best...


Again John, you are my favorite scientist when is come down to the nuts and bolts. Your comments are straight ahead and make sense in a way that I really enjoy. Thanks for the input. I learn alot from you.

[JonFairhurst]

Thanks for the kind words!

Aeons ago I majored in electrical engineering and computer science with a focus on digital signal processing. DSP wasn't just a whim. I had hand built an analog compressor for my guitar as a high school junior. My electives in college were pretty much all music courses.

After school all of my opportunities were related to image processing, rather than sound processing, but it's pretty much the same stuff, just in 2-d rather than 1-d.

I now work for a Japanese company, and in an international standards group. Communicating clearly is an important part of my job.

But the best way to develop communication skills? Write a few thousand posts on an open forum!

All the best...

[Larry Seyer]

For me it depends on what the format of the final mix is going to be.

If it's a CD project, it has always sounded better to me to record, mix, master, deliver at 44k1. I usually try to avoid sample rate conversion whenver possible.

However, if there is an analog step between the multi-tracks and the CD master, then 48k is an acceptable alternative. Analog transfers seem to 'fix' the 44k1 / 48k issues that I've had in the past.

Analog mastering is always preferred when mutli-tracks are done at 48k.... Even if I bounce a mix file to 48k/32bits, the analog mastering step makes the transition to 44k1 sound smooth to me.

No doubt about it though... 88.2 or 96k sounds better for those projects where quality is a priority...

But if the end product is going to be listened to on Itunes, what are we wasting our processing power on? Some future generation?

I'll go along with the 'do it for the Art' argument for a little while, but for my ears, 44k1 is plenty.

Bit depth is another matter entirely however.

I do everything at 32 bit float...

Especially now since everything get's squashed with compression to the brink of death, those extra bits make a huge difference in how much you can squeeze out of a mix.

Just my 2 cents.

Larry Seyer

[JonFairhurst]

...I do everything at 32 bit float...32-bit float must be especially important when creating impulses. The ratio from tip to tail can be HUGE.

Great point about using analog to convert sample rates. It introduces some noise, but the noise is neutral, natural and smooth. Digital quantizing noise can be biased and rough - especially if you aren't using enough bits.

The key to a good analog transfer is good converters. If you put out synchronous clock noise at 48 kHz and it gets through the filter and sampled at 44.1, you'll get energy at 3.9 kHz (the difference frequency). With good converters/filters, synchronous clock noise isn't the slightest of problems.

The other key is to have a good, rock-stable clock. Clock phase errors will lead to sample noise on analog transfers.

But regardless of the theory - if it sounds good, it is good!

[alanb]

if it sounds good, it is good! Duke's words really can be applied properly to anything, even synchronous clock noise... I love it... )(~

[JonFairhurst]

Though largely unknown, Duke Ellington was a giant in the DSP world... %-

[RickD]

Even if you recorded at 44.1, you won't have a direct 44.1 throughput if you're using a sampler. Samplers change sample playback frequency to interpolate notes and deal with tuning.

I've never done any serious testing, but I've always had the feeling that samplers trash audio (to a greater or lesser extent). Some playback plug-ins, like Intakt, severely mulch audio. Running audio through any sampler software sucks some of the life out of it.

Lee Blaske

It's nice to see I'm not the only one who thought this.

[Steve_Karl]

At the end of my work sessions I've been capturng the mix to a stereo file off of the outs of my Soundcraft 400B into a Gina24 and then a SoundForge 24/96 file for about a month now. I did one yesterday at 24/48.
I can hear the difference and preffer the 24/96 format. There's more detail in the highs.

[Larry Seyer]

This thread is getting off track, and I think it's possible some people are going to be confused.

We're talking about two different things. The sample rate of the session is one thing. The sample rate of a sample library is another thing. Samples in libraries are going to be pitch changed in various ways (retuned so that a single sample covers multiple notes, pitch bending, vibrato, etc.). The result will resampled, so there's no escaping some digital mulching in the process. Since this will be happening, the difference between 44.1k and 48k for the sample library is fairly moot. A 44.1k sample library is not especially advantageous for CD work, and a 48k library is not especially advantageous for video work. Actually, 48k should be marginally better for both applications. In neither case will the sampler be playing back the content into your session untouched.

Regarding the overall sample rate of the session, if it's going to CD 44.1 or 88.2 is better, and if it's going to be video, 48k is better for today's playback equipment.

For clarity's sake, perhaps this thread should be kept JUST about sample library sample rates, and a new thread about session rates should be started.

Lee Blaske

Lee,

I'm not trying to hijack this thread... I think sample rates matters in both cases.

If you're going to be adding any 'live' instruments or vocals to a sample based recording, then the format chosen for the recording does matter. (which is why I mentioned it).

Sample libraries will do a 'sample rate convert' on the fly for those notes that are playing back at sample rates / pitches that do not match the rates/pitch they were recorded at. (samplers are ALWAYS a compromise compared to 'live' recordings)

But if the sample library was sampled at 44k1 and there is no pitch shifting going on, and the notes are not being 'stretched', then there will be a large percentage of sounds coming from the sampler that are not changed... lessening the ill effects of the sample rate / pitch change process of the sampler. (assuming a CD project that is).

Some of the work that I did for GigaStudio was done at 48k and some of it was done at 44k1.

Usually the 48k stuff was done that way because it was generally assumed that the people using those functions/sounds would be using them in film/video/post. (i.e. GigaPulse Impulses etc).

But most of the sample libraries that I develop are done at 44k1.

This is because I 'assume' that the library is going to be used more for creating CD's than it is going to be used for creating film/video/post works. (this is purely a guess on my part... I have no way to really predict WHAT way a sample library will be used).

My point is the less times / notes that a sampler has to do pitch shifting / sample rate converting the better. But since samplers are going to be the weakest link in a recording anyway, it is my opinion that they should be the ones to make compromises with regards to sample rates.

Instead of doing 88k2 or 96k recordings for sample libraries, I do 44k1 versions. It halves the amount of work that the sampler has to do during playback and the difference one hears between a library recorded at 44k1 and 88k2 is not worth giving up half of the number of notes that can be played by the sampler.

Those who wish to do projects at 96k or 88k2 can do so making the sampler do the pitch change and having the 'live' instruments be recorded at the desired higher sample rates.

IMHO, this is a more efficient way to work and a better use of samplers and sample rates.

Best to you!

Larry Seyer

[howardv]

Those who wish to do projects at 96k or 88k2 can do so making the sampler do the pitch change and having the 'live' instruments be recorded at the desired higher sample rates.What I usually end up doing to not overwork gs3 (and I've never been able to get ni to upsample on the fly) is freeze the sampe track, upsample it, then move it over to an 88.2k audio project to record live instruments and vocals. Which isn't that inconvenient if the sampled tracks get laid down 1st. But when I want to add some sampled midi instruments later in the project, then I do a downsampled submix, freeze the new tracks, upsample the new tracks, and finally move them back to the 88.2k project trying not to mess up the alignment. Much simpler working with hi-res libraries and one sample rate throughout. Like the Piano West library which pushes processing and memory to the limit but doesn't give up any notes. I'm thinking with the new 64-bit dual cores that processor and memory limits won't be such an issue anymore... as soon as the sample engines catch up. Which when they do, will make sampling a library today at 44.1k look pretty short sighted. When sampling at hi-res and just making down-sampled libs for the market of the moment would be so easy.

Howard

[gregjazz]

Maybe it's just me, but I find 88.2k a much more logical samplerate to use other than 48k or 96k for making CDs. This is because when you convert it to 44.1k it divides evenly, so you get a lot fewer artifacts in the audio. (Mind you, I'm talking about projects that are eventually going to end up on an audio CD)

[pointybird]

Gigastudio for example seems to play back 44k libraries particularly badly at 48k.

I do mainly film work, so I used to stay in 48k but then I did various tests with VSL samples, and Gigastudio (and GVI)'s output has some horrible artifacts at 48k (with 44k samples). Thsi is particularly apparent on percussive low frequency sounds: for example try timpani (or piano) and boost the top end a lot (eq) - it sounds like an early Casio sampler.

So now I stay in 44k and convert with Barbabatch to 48k for delivery. The weird thing is that Logic's EXS24 does the same thing, with the same sounds MUCH more cleanly. In fact I don't notice any artifacts with EXS24.

Dom

[RickD]

How the hell did this old thread come up? i swear it was on the front page.

[ohernie]

I think it's forces of evil spamming the forum :p.

Somebody spams the thread, the forum software sees it as a new post and "promotes" the thread to the frontpage, a moderator deletes the spam and we scratch our heads.

Ernie