sample rates?

[edmann]

Greetings Charles

Do you have any thoughts or experiences to share regarding the of various sampling rates?

thanks!

Ed Mann

[Charles Dye]

Ed,

I've used 192k and 96k, but for the time being most of my work is still done in 44.1/48k. The main reasons are that when I work in pop or latin music the producers need the higher track counts, and when I mix whether rock, pop or latin I want the more efficient DSP usage. This is not to say I don't want to go to 96/192k and I intend on it once I have all Accel Processor cards in my chasis.

Eric Schilling, a very talented engineer who mentored me as a young assitant, is always very forward thinking with his early adoption of new technology. (He introduced me to DAW's in 1991.) He's been recording everything @ 96 or 192k for years now (prior to PT HD--he used Fairlight), with the intent that one day the records may in the future be remixed for a 96 or 192k consumer format.

Anybody else? Where are you with sample rates? What is your reasoning to use the higher rates or stick with 44.1/48k for now?

[jbuntz]

Sample Rates higher than 96 supposedly increase aliasing errors and actually reduce fidelity.

[Noah]

Let me pull out my Soapbox here...

This is from a plug-in developer so here goes...

While many don't hear a substantial difference in 44 vs. 192 for source audio, it becomes very apparent when doing any kind of sonic manipulation. When we were doing the Distortion mode for our new plugs we started fooling around with different sampling rates, and how much up sampling we wanted to do. I could hear a huge difference in distortion quality with lower sampling rates vs. higher ones. (High being much better) This difference was so great that when Ken turned it up in his lab I could tell the difference every time double blind from outside the lab at my desk 20ft away.

Really nice saturation and distortion in plugs is only going to come with higher sampling rates. 192 is just the start here (we were up sampling 4 times this and I could here the difference)...While it might not matter so much in the playback of this audio, it is extremely important when you are manipulating/mangling it. My feeling is once the bar is raised people will start to say "Hey this EQ, or this Compressor, or this Filter really sounds amazing!" but it ain't going to happen while at 44k.

Ok, I will get off my soapbox now...

Noah

[Noah]

Plug-ins can do a fair amount of their own up-sampling to get a higher sampling count. But I suppose it all depends on the quality of the up-sampling, and the respective CPU hit. When we look at process usage of our plugs, (if we are doing up-sampling) it can take up a fair amount of resources. If you are running 3 effects plugs on a track this ends up being pretty considerable if you need to up-sample then down-sample the audio 3 or 4 times. Plus there is room for error as you repeat the process. It all depends on how good you are/how much you know about writing your own algorithms. Some are better than others and you will hear that difference...

If people are interested we can keep going into it further. I can also ask Ken or Bob to pull in and write more about the technical aspects of why we need to up-sample for specific processes.

Noah

[Charles Dye]

Noah,

Thanks for the great info + yes please, I (and I'm sure others) would be very interested to hear more about the technology behind the plugs many of us rely on for their great sound. It helps us make decisions about how to use them when we work.

Thanks.

[djui5]

Charles,
Eric Schilling is a great engineer. I worked at Transcon for a while but never had the pleasure to work with him on a session. I'll bet you learned a lot from him...

(Not trying to hyjack your thread)

[sound and form]

Hi,

As a plug-in developer too, I can fully endorse Noah's sentiments about higher sample rates.

This reminded me of a thread from the DUC. Here, Frederick Umminger over at Digi discusses some of the real benefits of higher sample rates when dealing with audio processing:

Steven Massey
sound and form

[Charles Dye]

Randy,

Yes, Eric is indeed brilliant. And I learned tons. Sitting behind him listening to him mix was the best training my ears ever received. The way Eric sculpts his low end was a big influence. We still see each other fairly regularly. Great guy too.

Susan is the best as well. She managed Crescent Moon when I was an assistant there.

[Steve Smith]

I swear i like things like Compression at 96K as well.. the plug in quality jump wa more apparent to me than the actual sonic inprovement initally


but maybe I am a moron.. who knows?

[djui5]

Susan is awesome! She's the best studio manager I've known yet. I haven't talked to her in a few week's or so. Tell em I said hello if you see them please.
Also if I may ask a question, how do you feel about 88.2. There's some talk about multiples of 44.1 and the math involved in downsampling that relates to multiples of 44.1. Have you done any testing as to which is better...88.2 or 96? Thanks in advance for your time

[Steve Smith]

I swear i like things like Compression at 96K as well.. the plug in quality jump wa more apparent to me than the actual sonic inprovement initally


but maybe I am a moron.. who knows?

[Charles Dye]

Originally posted by djui5
How do you feel about 88.2. There's some talk about multiples of 44.1 and the math involved in downsampling that relates to multiples of 44.1. Have you done any testing as to which is better...88.2 or 96?

Randy,

I've done no tests, but logic tells me, for the sake of maintaining the accuracy of your mix, that working at higher multiples (88.2/176.4k) of the intended consumer sample rate (44.1k) and downsampling after digital mastering would be the best instead of using sample rate conversion.

OTOH, if you know you will be mastering analog then you might want to go for the higher sample rate.

Does everyone agree?

[Curve Dominant]

Does everyone agree?

A thread like this is never complete until Nika Aldrich chimes in.

I'll call Nika tomorrow and see if he has time to contribute to this discussion.

[edmann]

Charles -

going back to the pre-plugin phase of this thread:

Do you hear differences between your trax recorded at (for ex) 48k vs 96k?...and....96k vs 192k?

thanks man -

edmann

[edmann]

Originally posted by edmann
Charles -

going back to the pre-plugin phase of this thread:

Do you hear differences between your trax recorded at (for ex) 48k vs 96k?...and....96k vs 192k?

thanks man -

edmann

~bump~

:-)

[blackcatdigi]

Originally posted by Charles Dye
I've done no tests, but logic tells me, for the sake of maintaining the accuracy of your mix, that working at higher multiples (88.2/176.4k) of the intended consumer sample rate (44.1k) and downsampling after digital mastering would be the best instead of using sample rate conversion.
(snip)
Does everyone agree?

Yes and No.

Logically, the simpler math analogy makes sense. However, I did have an interesting discussion with the Benchmark guys about this, and they stated that in actuality, the even division (say, 88.2 down to 44.1) resulted in more aliasing than noneven calculations...

[edmann]

Originally posted by blackcatdigi
Yes and No.

Logically, the simpler math analogy makes sense. However, I did have an interesting discussion with the Benchmark guys about this, and they stated that in actuality, the even division (say, 88.2 down to 44.1) resulted in more aliasing than noneven calculations...

Nika discusses and illustrates the same thing - thanks for including this important point blackcatdigi

Nika! come in........(?)

[AUDIOSFX]

Film is really 16. 48k or even 16 44.1 dolby ac3 encodes 16 48 but is variable based on sonic necessity(not constant )both consumer and commercial

If the end result will be high resolution..i.e client request then 96 or above makes sense. The problem is the client base does not request high resolution. Especially when mp3 resolution is crushing all our high resolution goals.

I do a great deal of multichannel mixing and have the capability of 192.. The problem is encoding masters at High resolutions. DVD authoring encoders like DTS and or AC3 do not support high sample rates...yet..DCD is not realisticly applicable and only Minnetonka makes a pcm DVD encoding solution at a reasonable price. Sonic Solutions is in the game but most of these encoders are Pee Cee only. ANd sonics is not cheap even for a bigger facility.

so usually 24 48music..
16 48 film..

-ws

[djui5]

Originally posted by blackcatdigi
Yes and No.

Logically, the simpler math analogy makes sense. However, I did have an interesting discussion with the Benchmark guys about this, and they stated that in actuality, the even division (say, 88.2 down to 44.1) resulted in more aliasing than noneven calculations...

Thanks for posting.

How is the aliasing in comparison to 96? More is less? Thanks a ton for your answers...I really appreciate it.

[Charles Dye]

Originally posted by blackcatdigi
I did have an interesting discussion with the Benchmark guys about this, and they stated that in actuality, the even division (say, 88.2 down to 44.1) resulted in more aliasing than noneven calculations...

Casey,

I'm glad I asked. Do you have any more information on this? Again, it really doesn't make any sense. How could that be? But I have to say I'm not necessarily surprised. What do the Benchmark guys suggest we do instead?

Thanks.

[blackcatdigi]

Well Charles, I must say it was a rather informal conversation with John and Allen on the floor at NAMM, and the topic just happened to come up... The only reason I even remember it is that I was so shocked to be told this.

Hopefully, one of our resident 'Rocket Scientist' will jump in and confirm or deny what I was told. Maybe Atticus can get the Benchmark line on this for us...

IME, as analog is always a part of my chain, I always (re)convert to the desired SR. So, I've not done any testing of this purported theory.

But the line of thought discussed was:
Later generation SRC algos have progressed to the point of being less degrading than the aliasing of even divisioned downsampling...

My interpretation of that info is "don't use even divisibles..."

Smart guys, take it...

[Charles Dye]

Man, I'm still stumped by this one. How could an algorithm do a better job of altering a sample rate than dividing it by two? When you divide something by 2.176871 (converting 96 to 44.1k) aren't there going to be remainders that will be rounded + therefore make the result less accurate than by even division?

I'll see if I can get a more qualified answer than my original post.

[Nika Aldrich]

[QUOTE]Originally posted by Noah
Let me pull out my Soapbox here...

I just dusted mine off as well...

While many don't hear a substantial difference in 44 vs. 192 for source audio, it becomes very apparent when doing any kind of sonic manipulation.

Not exactly "any." It becomes apparent with certain types of non-linear processing, the degree to which is determined by the type and degree of non-linearity involved. Certainly distortion modelling is a great example of the benefits of high sample rate processing though this does not necessarily speak to the benefits of high sample rate recording. Further, some processes are disadvantaged with higher sample rates.

When we were doing the Distortion mode for our new plugs we started fooling around with different sampling rates, and how much up sampling we wanted to do.

Precisely. This speaks entirely to the issue of the importance of processing at the higher rates, though upsampling is a perfectly adequate way to do so, and in fact can reduce distortions that can happen otherwise.

Really nice saturation and distortion in plugs is only going to come with higher sampling rates.

Not necessarily. Manufacturers just have to do upsampling, which does not necessarily require that the session operate at the higher rate. There are several processes that do so already, and there are plugins in Protools, for example, that upsample the material to 384KS/s or higher prior to the actual process occurring.

Nika.

[Nika Aldrich]

Originally posted by mdbeh
Interesting post, Noah. You mentioned up-sampling--how much of the benefit of processing at higher sampling rates is dependant on recording at those higher rates, and how much can be gained simply by up-sampling material that was recorded at 44.1 or 48?

This is a great question. The benefits he is speaking of can all be accomplished through good quality up and down sampling as opposed to actually recording at the higher rates.

There is an argument that too much up/downsampling also has its consequences, but there is an accompanying argument that says that recording at 96KS/s inherently adds more distortion in certain undesireable areas.

Nika.

[Nika Aldrich]

[QUOTE]Originally posted by Steve Smith
I swear i like things like Compression at 96K as well.. the plug in quality jump wa more apparent to me than the actual sonic inprovement initally

There are ways to design compressors so that this shouldn't be a factor, but it is likely you were using ones that were not designed this way. Compression, however, being an inherently non-linear process does fall into the category of processes that could take advantage of high rate processing.


but maybe I am a moron.. who knows?

Nope - just the bright red chicken tikka masala acting on you.

Nika.

[Nika Aldrich]

Originally posted by Charles Dye
Man, I'm still stumped by this one. How could an algorithm do a better job of altering a sample rate than dividing it by two? When you divide something by 2.176871 (converting 96 to 44.1k) aren't there going to be remainders that will be rounded + therefore make the result less accurate than by even division?

Charles,

There are essentially two types of sample rate converters at play in our industry right now. Asynchronous converters are ones that happen in hardware boxes outside of a piece of equipment wherein two boxes each slaved to asynchronous clocks are then reconciled and the sample data is modified.

The second type of sample rate converter is the type that happens inside the computer where there is no clock - a piece of data is merely modified so that it can be replayed at a different clock rate and sound the same.

This type of process would logically be done in the way in which you describe. In actuality it is done by taking the "incoming" clock rate and upsampling the data to the lowest common multiple of the incoming and outgoing clock rates. That means that for 44.1k and 48k the data is upsampled to something like 35MHz. Once it is upsampled to this rate it is then downsampled to the appropriate sample rate. Here is the catch, though. In order to do this all less expensively and with less processing involved, most software programs upsample to the same rate regardless of what they are coming from or going to. Therefore, whether the "inbound" rate is 44.1, 48, 88.2, 96, 176.4, or 192, the SRC takes the material and upsamples it to 35MS/s or so. Then it can downsample to any of those destination clock rates using a very simple formula. This means of doing it means that manufacturers only have to come up with a single set of filter coefficients for each sample rate. Further, there aren't any disadvantages to doing it this way (though I would have to explain that more in depth).

Therefore, it doesn't matter what the "from" sample rate is - they all get upsampled to 35MHz or so, removing any potentially perceived advantage of a direct-integer based downsampling like 88.2KS/s to 44.1KS/s. In essence, nobody does direct-integer based downsampling anymore because they don't need to.

I hope this helps?

Nika.

[studjo]

Thanks Nika

this is the first time I think I understand this whole "samplerate doesn't matter" thing.

Jo

[AUDIOSFX]

Brilliant Nika, and thank you for the information. Just curious if this is the case. How much influence does the external clocking device have on incoming signal and if so whos got the best clock? since its being reconciled at a rate far above its clocking source?

-ws

[Nika Aldrich]

Originally posted by AUDIOSFX
Brilliant Nika, and thank you for the information. Just curious if this is the case. How much influence does the external clocking device have on incoming signal and if so whos got the best clock? since its being reconciled at a rate far above its clocking source?

-ws

Absolutely none. Since the process is done offline there is no clocking concern in any capacity whatsoever.

Try this - what is:

1 + 1?

How about:

1+ 1?

Or:

1 +1?

Or even:

1




+



1?

It does not matter how erratic the information comes in, the results of the math are the same in digital processes.

The only places that clocks make a difference in the studio are at the A/D, D/A and aSRC converters (that's "asynchronous" SRCs). Nowhere else in a studio, including offline processes or internal sample rate converters, does the clock make a difference.

Now in aSRCs there are definite clock issues.

Nika.