The 96k difference

[Archmart]

Hey Hey!

Help me out?

I'm about to start a project and I've usually worked at 24bit/44.1, but I'm thinking of upping the ante to 24bit/96k.

So what difference am I looking at here, in terms of increased latency (a little over double?) and decreased instances of plugs (a little less than half what I'm used to?) But latency compensation plugs will still work without extra math, yes? (though different for the Precision Plugs)

It's a fairly jazzy, folky, bluegrass pop sort of thing, so I'm thinking the higher sampling rate is going to be a little more useful than some other stuff I've done and, heck, drive space is cheap, my G5 has never broken a sweat, etc.

Still, perhaps I'll regret it after I get going?

I'll be tracking largely on a ProTools HD-3 system although I may need to do some remote things on a Digi 002R. I'll be editing and mixing at home with the 002R.

Archmart

 

[Stacy]

Personally I have never tracked over 48khz. I think anything over that is just a waste of drive space and will only bogg things down. I am however a real believer in the higher bit rates. That seems to make alot more sense to me than the sampling rates. And the mastering sessions I have sat in on are what have shaped my oppinions. I have heard the A/B comparisons and really never personally heard a compelling argument for the 96khz thing. I'm sure you are going to have a number of audiophiles on here who will adamantly disagree but I would be most curious to know where they got their oppinions other than what a stat sheet says or what some \"Renowned\" engineer wrote in an article once.

 

[Archmart]

Hey Hey!

Certainly the jump from 16 to 24 bit is a much bigger deal than the sample rate.

And yes, I'd agree with you about listening to A/B comparisons. Wow does it take a serious system and the right material to easily tell the difference above 44.1.

But I've been thinking more lately about more long term questions. What am I missing that I don't notice right away but over time I end up training my ears/brain to ignore? Someone (I'm forgetting who, either Moog or Neve... or it might've been someone else with a four letter name) pointed out that we can tell the difference between different shaped waves at the same frequencies, so we must be hearing harmonics way up there. It just makes sense to me that we'd have a tendency to do an A/B and not hear enough of a difference to warrant the trouble, but inevitably years later we'll be appalled at what we used to listen to. While that's inevitable to a certain extent, I'm certainly interested in avoiding it as much as reasonable.

So I'm thinking 96k. Now... that being said, I might actually record it at 88.2k (easier math and all) and immediately dumb it down (in half) so it's more workable while at least knowing that I've got the original takes at a higher quality for posterity.

Archmart

p.s. Where's that tiled tunnel? And how does it sound?

 

[Fundy]

I think the irony is the actual latency in the ASIO drivers is cut at a higher sample rate. Although the system load is increased as a payload.

With higher sample rates, I don't think there is any real benefit unless you record at those sample rates to start of with.

 

[bedhoe]

There is a huge difference in sound quality when working in 88khz compared to 44!

I don't understand at all when people say they can't hear the difference.

The UAD-latency is cut in half, and it is a hell of a lot easier to mix well in 88khz (at least once you've gotten used to it).

Yes most UAD plugins use more DSP but some don't really change because they upsample internally anyway.

The stereo image is much clearer and you'll find it is easier to decide what to cut out and eq when you work in 88khz.

All this might be different with lesser converters though. I haven't really tried anything but apogee.

Just make sure you downsample it with good software (voxengo r8brain pro is a good one)

 

[Dan Duskin]

88.2 is pretty safe... and if you run out of CPU/DSP you can bounce and freeze, and even then convert to 44.1kHz for additional processing (with high quality upsampling EQ's from UA like the Neve's, Pultec & PreEQ).

 

[Stacy]

Hey Hey!

Certainly the jump from 16 to 24 bit is a much bigger deal than the sample rate.

And yes, I'd agree with you about listening to A/B comparisons. Wow does it take a serious system and the right material to easily tell the difference above 44.1.

But I've been thinking more lately about more long term questions. What am I missing that I don't notice right away but over time I end up training my ears/brain to ignore? Someone (I'm forgetting who, either Moog or Neve... or it might've been someone else with a four letter name) pointed out that we can tell the difference between different shaped waves at the same frequencies, so we must be hearing harmonics way up there. It just makes sense to me that we'd have a tendency to do an A/B and not hear enough of a difference to warrant the trouble, but inevitably years later we'll be appalled at what we used to listen to. While that's inevitable to a certain extent, I'm certainly interested in avoiding it as much as reasonable.

So I'm thinking 96k. Now... that being said, I might actually record it at 88.2k (easier math and all) and immediately dumb it down (in half) so it's more workable while at least knowing that I've got the original takes at a higher quality for posterity.

Archmart

p.s. Where's that tiled tunnel? And how does it sound?

Good luck with the experiments. I think you have a good point in that we are used to hearing digitial music in its early days sounding so frail and bright. But then that always seemed to be an A/D - D/A converter issue more than anything else.
The tunnel sounds great!:

 

[neil wilkes]

I do not understand why it is constantly being claimed that 88.2 is \"easier math\". Easier for what, please?
It's not SRC, because all the decent SRC use whole numbers to do the conversion anyway - no fractions, no rounding required. Simply use the GCD and work it all out for yourself. Modern SRC go up, then down to avoid fractions - for example:
96K to 44.1K
GCD for these is 300 (as for all sample rate conversions) so we do this:
96000/300 = 320
44100/300 = 147
So from 96K to 44.1 we first go up by a factor of 147, and down by 320.
96000*147=14112000/320=44100. QED.

The best argument for using High sample rates is not the frequencuy response, unless you're mixing opera for dogs & bats. It's all about the quantization in the converters.
To quote from a very well respected friend of mine who really understands this stuff:

96 kHz or 192 kHz is a frequency of quantization of analog audio signal by ADC, it is not the frequency of real audio signal, produce by some instrument or whatever. So if you consider CD, mp3, AC3 or DTS with 44.1 or 48 kHz, you will have only 2-3 quantizes for the whole period of high frequency signal (15 - 20 kHz) ! It is enormous distortion of high frequency signal in ADC - DAC process. Some improvements can be made by integrating in DAC some intelligent interpolation scheme to approximate the jogged line of high frequencies to real sinusoid (if you consider fixed frequency, say 15 kHz that will be represented as jogged line by DAC, not sinusoid). Considering 192 kHz process, that will give you 9 - 14 quantizes for the period of high frequency audio signal, that will give much improvement in DAC processing of high frequencies.

And, when I was asking for more detail, I got this reply

...considering fixed frequency (say 15 kHz) and quantizing it with different rates (32, 44.1, 48, 96, 192 kHz etc.) will represent different quality on the output in comparison with the original. If you have fixed frequency, it's spectrum will be Delta-function. But if you process it through ADC-DAC with different quantization frequencies, you will have a lot of additional parasitic harmonics with different amplitudes (the main amplitude on 15 kHz will obviously be reduced), and the spectrum of output signal will be very far from Delta-function. Because the output signal is not sinusoid but is still jagged line. And the higher the quantization frequency the closer the output signal will be to sinusoid and spectrum to Delta-function.

(we went through this argument about 192 Stereo, as my original thinking was that there is no point because consumer equipment cannot reproduce a 96KHz frequency, neither will most amplifiers)

Hope this helps.

In short, 96KHz is well worth doing.

 

[Schaap]

Thanks Neil for explaining this in such a nutshell :lol:

Henk

 

[daverich]

I find the main reason 88.2khz sounds better is that plugins sound nicer, not that the actually audio sounds much different.

Kind regards

Dave Rich.

 

[Dan Duskin]

I find the main reason 88.2khz sounds better is that plugins sound nicer, not that the actually audio sounds much different.

And that is exactly the difference you will notice the most. I wish this was better understood in the industry.

 

[entoine]

I find the main reason 88.2khz sounds better is that plugins sound nicer, not that the actually audio sounds much different.

So that make sense that plugins upsample internaly like pultec, neves and precision Eq ones...

but do we have to RECORD at 88Khz... ? In a DVD about mastering, the guy was saying that frequencies above 20Khz are subject to intermodulation and create frequencies below 20Khz (not in the file but in the ear I guess)... it make sense. And for him the \"artefacts\" in the high frequencies are the difference we ear between 48Khz and 96Khz.

 

[Paul Woodlock]

To quote from a very well respected friend of mine who really understands this stuff:

96 kHz or 192 kHz is a frequency of quantization of analog audio signal by ADC, it is not the frequency of real audio signal, produce by some instrument or whatever. So if you consider CD, mp3, AC3 or DTS with 44.1 or 48 kHz, you will have only 2-3 quantizes for the whole period of high frequency signal (15 - 20 kHz) ! It is enormous distortion of high frequency signal in ADC - DAC process. Some improvements can be made by integrating in DAC some intelligent interpolation scheme to approximate the jogged line of high frequencies to real sinusoid (if you consider fixed frequency, say 15 kHz that will be represented as jogged line by DAC, not sinusoid). Considering 192 kHz process, that will give you 9 - 14 quantizes for the period of high frequency audio signal, that will give much improvement in DAC processing of high frequencies.

And, when I was asking for more detail, I got this reply





[quote:8yyohwko]...considering fixed frequency (say 15 kHz) and quantizing it with different rates (32, 44.1, 48, 96, 192 kHz etc.) will represent different quality on the output in comparison with the original. If you have fixed frequency, it's spectrum will be Delta-function. But if you process it through ADC-DAC with different quantization frequencies, you will have a lot of additional parasitic harmonics with different amplitudes (the main amplitude on 15 kHz will obviously be reduced), and the spectrum of output signal will be very far from Delta-function. Because the output signal is not sinusoid but is still jagged line. And the higher the quantization frequency the closer the output signal will be to sinusoid and spectrum to Delta-function.

.....[/quote:8yyohwko]

Whether the sine wave was sampled 2 times or 1,000 times along a single cycle, once you've filtered out the 'jagged line' the output will be perfect 15kHz sine wave. You only need 2 samples per cycle ot accurately reproduce a sine wave. Hence the nyquist theory.

Sampling a 15kHz sine wave makes no difference whether you do it at 32kHz or 192kHz

Reading your two quotes above I suspect your 'friend' doesn't actually understand it that well at all.

 

[Paul Woodlock]

Thanks Neil for explaining this in such a nutshell :lol:

Henk

Except it's wrong.

 

[Trace]

Thanks Neil for explaining this in such a nutshell :lol:

Henk

Except it's wrong.

That Said Paul, what do you record at and why?

Thanx TRACE

 

[Matt Hepworth]

Everyone should read this white paper: http://www.lavryengineering.com/documen ... Theory.pdf

 

[joenovice]

I've read that before... and I just read it again.

What I don't understand is why Dan puts down higher sampling rates but still produces products that sample at 96k?

 

[Paul Woodlock]

Thanks Neil for explaining this in such a nutshell :lol:

Henk

Except it's wrong.

That Said Paul, what do you record at and why?

Thanx TRACE

I record at 44.1/24

Why? because I don't have enough horsepower to bother with anything higher to take advantage of any slight increases in qulaity of plugin processing.


Neil's friend is still wrong though.

 

[bulls hit]

I've read that before... and I just read it again.

What I don't understand is why Dan puts down higher sampling rates but still produces products that sample at 96k?

I think it has more to do with it being easier to build anti aliasing filters when sampling at higher rates - the slope doesn't need to be so steep.

As Paul Woodlock points out, Nyquist demonstrated that you can recreate an analog waveform from digital samples with 100% accuracy as long as the sampling rate is at least twice the highest frequency in the original waveform. If 20KHz is the highest we can hear, then sampling at 40KHz is enough. 44KHz became the standard sampling rate to allow some room for the filter to cutoff all the aliasing nasties above 22KHz. However to go from full noise at 20Kz to -infinity at 22KHz requires a very steep, and relatively expensive filter when doing DA. Some filters compromise, and start cutting at maybe 16KHz, so you lose some of the high end clarity. The alternative is to use a higher sampling rate, and nice easy filter. That's how I read it anyway.

 

[Macc]

Whether the sine wave was sampled 2 times or 1,000 times along a single cycle, once you've filtered out the 'jagged line' the output will be perfect 15kHz sine wave. You only need 2 samples per cycle ot accurately reproduce a sine wave. Hence the nyquist theory.

Sampling a 15kHz sine wave makes no difference whether you do it at 32kHz or 192kHz

Reading your two quotes above I suspect your 'friend' doesn't actually understand it that well at all.

Absolutely totally correct Paul... I literally started typing that response three times yesterday but chickened out. A DA convertor's output is NOT a jagged 'digital' stepped line, it's an analog waveform. The poster's mate is talking bollocks.

One very common misunderstanding on this point is that people generally seem to think that if you have only two points, it will give you a triangle wave, that's the simplest way, right, a straight line? It's the 'common sense' point of view. However physics doesn't really see it like that. A triangle wave is, counter-intuitively perhaps, much more complex (it has harmonics etc). In physics, you don't get any more simple than a sine wave - simple harmonic motion anyone? And so two points is enough to describe it, and have it reconstructed perfectly at output from a good DA convertor.