Linking two computers

urban_productions · Post by **urban_productions** » Sun Aug 15, 2004 3:25 am

On 2004-08-10 16:38, cleanbluesky wrote:
My destinations get mixed down to 44.1k I like recording at 96k as I believe that a person should record in the best that is available and keep the masters that way. I wish that I could record in 192k or even DSD (sigh) but SCOPE doesn't support 192k and I'm not even sure if there is any software to put effects on DSD. I believe mastering software for DSD is recent...

we made listening tests and the converters currently available can not handle 96KHz es well as 48... the result was : all students said 48 sounds better.
dont trust numbers, trust your ears.

urban_productions · Post by **urban_productions** » Sun Aug 15, 2004 3:36 am

On 2004-08-13 10:57, astroman wrote:
it's an overstressed example by me ...

anyways the dsp´s on current cw-boards are outdated models. texas instruments makes 10x more GFOPS... would be nice to have those.

cleanbluesky · Post by **cleanbluesky** » Sun Aug 15, 2004 4:45 am

The converters available today - So you tested all converters, or just some high-end ones? Were you sure all were functioning properly?
Were the final recorded pieces analysed under the computer to see what the physical difference was (so you could find out why the 96k converters 'connt handle' 96k)?
Was the test double-blind, placebo controlled?
All students where?
Was any extra processing done to the audio streams?
Was the same hardware used to produce both audio streams (running first at 48k then at 96k)?

astroman · Post by **astroman** » Sun Aug 15, 2004 6:30 am

On 2004-08-15 04:36, urban_productions wrote:
... texas instruments makes 10x more GFOPS... would be nice to have those.

don't trust numbers...

the 'overstressed' example on the previous page shows in all details, how such stuff is engineered and how much differenve exists between theory and real world - not exactly entertaining, though...

One advantage of the Sharc DSP is it's outstanding library for audio purpose - it doesn't matter how many GFLOPS one can perform if the desired processing algorithm doesn't exist.

Of course it would be nice to have 2-4 times faster chips onboard (as those exist in affordable versions) - all other stuff is either too expensive (if it has 10 times the performance, but costs 20 times as much...) or is targeted at different processing tasks.

You just cannot write a (basic audio) DSP lib in a year - consider this a 5 to 10 man-years project IF you have EXCELLENT coders on staff. That's roughly a million Euro in wages alone, plus facility and gear costs.

These costs arise BEFORE you even know if you'll be able to deliver to market - and they will keep running while you (hopefully) succeeded.
Now go figure out what qty you have to sell in what time unit to not get broke

On the other hand there's not much need for outstanding speed in audio DSP processing - the biggest disadvantage of the current system design is the use of mainboard scratch memory, accessed via the PCI bus.

Other 'interesting' processings (like convolution) aren't impossible due to raw processing limitations, but to a different data flow design.

cheers, Tom

cleanbluesky · Post by **cleanbluesky** » Sun Aug 15, 2004 10:18 am

If scope were to have convolution reverb then it would be perfect... (that and a few extra SHARCs minus a few euros in price)...

Why can't SHARCs be used for convolution?

astroman · Post by **astroman** » Sun Aug 15, 2004 11:44 am

well, of course you can code the convolution on a Sharc DSP, you probably can do it even very good - it's just that SFP's current architecture is focused on real time processing and convoution with it's huge arrays doesn't fit here.
Your 96k question already revealed how tight timing margins for processing are.

But convolution isn't limited to reverbs, the Neuron synth uses a similiar strategy (wavelet transformation) for storing it's 'acoustic models'.

Wavelets have nothing to do with wavetables or wave playback, despite of the name - it's the name of a math process.
According to the Neuron's main programmer (S. Sprenger) it's one (new) possibility for audio encoding, but by far not the only one thinkable

There's a lot of stuff still to be revealed, but honestly - yesterday I fiddled around with a simple Flexor factory patch and was blown away by both the amount of sound variation possible and it's quality.
We do have a lot of great stuff on SFP which certainly isn't exploited yet

cheers, Tom

cleanbluesky · Post by **cleanbluesky** » Sun Aug 15, 2004 6:18 pm

Yeah, I think the reason that SFP is so good is that almost EVERYTHING on it is very high quality. The synths are more of a gift for electronic music but I would love some convolution...
I think that convolution is brilliant, although my only regret is that it is HARD to get some of the reverb that I like. There is one in Glastonbury, England in the ruins of an Abbey - there is a dome shape structure called the Friar's kitchen and I fell in love with the amazing reverb in there. Only trouble is I don't live there and I am not sure that the monks that tend the place would let me go in there with equipment anyway...
But I have seen some convolution files for units such as Manley Massive Passive, which can come in handy.
Do I have it correct that the reason that SFP is not suited for convolution reverb is that the Creamware cards are designed for real-time processing?
Would a convolution reverb be possible? Would the performance be any good?
Why hasn't someone made one?

astroman · Post by **astroman** » Sun Aug 15, 2004 7:47 pm

On 2004-08-15 19:18, cleanbluesky wrote:
...Would a convolution reverb be possible? Would the performance be any good?
Why hasn't someone made one?

not in a reasonable design - you could do the actual calculation on the DSPs, but you'have to shift huge arrays of data over the PCI bus between the board and mobo memory.
That would block the card's control and result in a PCI bus overload - the infamous one

It's a limitation of the current board design (which is some years old, indeed) - and if CWA had the resources they'd be the first to come up with a redesign, including a large amount of onboard memory and new Sharcs.
If we contribute as customers we might be able to see this project succeed - without cash flow, forget it

cheers, Tom

ps: convolution is a (strategically) fairly simple process, which processes each value of an input array (the wave data) with each value of a 'control' array (i.e. the impuls response) by a certain math function.
A second of audio at 48k with an impulse response of 1 Mbyte means 48k x 1 million = 48 billion calculations per second, if the calculation can be done in one cycle.
If it's a complex calculation, multiply accordingly.
That's the brute force way - now the assemply wizzes are invited

The numbers above illustrate that there are NO high quality convolution verbs on PCs, that stuff definetely DOES NOT yet exist.
They process the attack phase of the sound and fake the tail usually

Post by **hubird** » Sun Aug 15, 2004 9:43 pm

cool!

clear as hell

cleanbluesky · Post by **cleanbluesky** » Mon Aug 16, 2004 12:36 pm

Personally I find convolution reverbs to be very fulfilling, and much easier and more natural than fiddling with a reverb simulation or reverb outboard (was considering buying TC Electronics M3000 until I discovered the joy of convolution... I also have TC Fireworx which I would not use for reverb unless I wanted a specific TC reverb sound)
So SIR convolution is not sincere? Are you sure?
What of Voxengo Pristine Space or Waves IR1 - both expensive and comprehensive... it would be unfortunate to find that they are 'fake' convolution?
Is there no other way that you could see it working other than 'faking' it?
I can understand that it takes a lot of processing power, if I load an 8 second convoution in 96k I am lucky to get a single channel of it!
That's why I would like to link two or more computers... still no reply on whether FX - Teleport is any good.

cleanbluesky · Post by **cleanbluesky** » Mon Aug 16, 2004 12:36 pm

Personally I find convolution reverbs to be very fulfilling, and much easier and more natural than fiddling with a reverb simulation or reverb outboard (was considering buying TC Electronics M3000 until I discovered the joy of convolution... I also have TC Fireworx which I would not use for reverb unless I wanted a specific TC reverb sound)
So SIR convolution is not sincere? Are you sure?
What of Voxengo Pristine Space or Waves IR1 - both expensive and comprehensive... it would be unfortunate to find that they are 'fake' convolution?
Is there no other way that you could see it working other than 'faking' it?
I can understand that it takes a lot of processing power, if I load an 8 second convoution in 96k I am lucky to get a single channel of it!
That's why I would like to link two or more computers... still no reply on whether FX - Teleport is any good.

AndreD · Post by **AndreD** » Mon Aug 16, 2004 12:54 pm

On 2004-08-16 13:36, cleanbluesky wrote:
still no reply on whether FX - Teleport is any good.

just get the demo

for me, it works fine with n2/sx2 but not with logic...

best,
andre

astroman · Post by **astroman** » Mon Aug 16, 2004 3:06 pm

On 2004-08-16 13:36, cleanbluesky wrote:
... So SIR convolution is not sincere? Are you sure?
What of Voxengo Pristine Space or Waves IR1 - both expensive and comprehensive... it would be unfortunate to find that they are 'fake' convolution? ...

oops, I should have been more precise - I actually referred to real-time application of convolution verbs.
Afaik that still requires special and very expensive hardware.

The offline application is something different, as you have a reduced dataset for 'live-preview' and once it's set up properly you render the stuff with as much time as necessary.

But be aware - I know convolution from image processing, where it has become a de-facto standard to squeeze a quality out of scans that just isn't there (unsharp-masking).
Those pictures literally shriek at my eyes and I'm anticipating the worst if that technology makes it's way into audio processing

Early implementations of digital audio were much appreciated (and admired) once they were new, but today we consider them more or less harsh and metallic

cheers, Tom

<font size=-1>[ This Message was edited by: astroman on 2004-08-16 16:08 ]</font>

cleanbluesky · Post by **cleanbluesky** » Tue Aug 17, 2004 12:27 pm

The analog versus digital argument is very interesting and considering that it is only measured in subjective terms, cannot ever be won or even an inch gained in the argument.
I didn't realise that there was a real-time convolution system... I read that TC Electronics latest reverb machines MAY travel down that line but I was talking about the off-line (as in pre-rendered or recorded impulse) type, which I adore. Voxengo do a wonderful program to create impulse files and the level of processing required can be seen there as it can take a long time for an impulse file to be rendered...
So do reverbs like IR1 and Pristine Space do a good job from a technical standpoint, in your opinion?

astroman · Post by **astroman** » Wed Aug 18, 2004 8:11 am

I must admit I don't have any opinion in this case - neither XP nor a VST host here

But Waves as a manufacturer is established and Voxengo got attractive prices...

Again there's a backward parallelism in statements about a lack of realism in alogrithmic reverbs if compared to convolution.

Wasn't it the same argument when sampling entered the market ?
Now, who wants a sampled Moog or Prophet ?
And those who can afford still prefer real orchestras - and honestly, no lib in the world can fake a violin virtuoso

I find it technically highly interesting, but I don't have no special need for realistic rooms.
But I totally agree with your enthusiasm about the architecture (and sound capabilities) of certain buildings - many created without even the idea of a calculator, let alone a computer

cheers, Tom

cleanbluesky · Post by **cleanbluesky** » Wed Aug 18, 2004 1:45 pm

My concept of convolution is that it will be the exact same reverb that you can capture, the only problem being that it shares the limitations of the digital medium in which it is used. Other than that I believe it is ace and produces a much more natural sound (IMHO) than even high-end simulated reverbs.

astroman · Post by **astroman** » Wed Aug 18, 2004 4:27 pm

well, it's just one way of doing things.

the high regard of quality convolution verbs is probably derived from the fact that you'd normally (certain SFP devices excluded

) have to spend a small fortune to 'simulate' realistic rooms.
So it must be good because it can keep up with xyz.

It is of course fascinating what a fairly simple math process can do, but finally it's nothing but a sound source.
One wouldn't want a surf-guitar verbed by an abbey's response, but prefer a shattering spring instead

Convolution adds big to the picture and I'm certain we've only seen the tip of the iceberg yet. The worst fact of convolution it's that's it's so boring unromantic and unemotional...

cheers, Tom

janila · Post by **janila** » Thu Aug 19, 2004 5:07 am

On 2004-08-18 14:45, cleanbluesky wrote:
My concept of convolution is that it will be the exact same reverb that you can capture, the only problem being that it shares the limitations of the digital medium in which it is used. Other than that I believe it is ace and produces a much more natural sound (IMHO) than even high-end simulated reverbs.

An analogy might be acoustic instruments and physical modeling synthesizers. A recording of an acoustic instrument shares the limitations of the recording medium, but it's still closer to the real thing than modeling. Maybe some day the modeling instruments and reverberation devices sound real to our ears but it isn't reality today. Pristine Space rules.

If you are convinced that 96k is a better format for you than 44.1k then you should try running 88.2k. The sample rate conversion from 96k to 44.1k is far more complex to do well than the conversion from 88.2k to 44.1k and the artifacts can be audible. If the end product is 48k (e.g. DVD) then 96k is a better choise than 88.2k. With Creamware cards you even get a few percent more processing power at 88.2k than at 96k.

cleanbluesky · Post by **cleanbluesky** » Thu Aug 19, 2004 1:48 pm

That is good advice janilia. I am getting another computer soon so linking the two up might take care of most of my problems. SCOPE already sorts most of my processor problems with regard to soft-synths anyways, all I care about is reverb, eq and amp-modelling convolution. The reverb and the amp-convolution take up a lot of CPU. I prefer to use mic'd up guitar amps just to be pristine and analog as possible but amp-sims are very, very good. I read a rumour that Perfect Circle's Mer De Noms was done almost exclusively with Line 6 Amp Farm.

astroman · Post by **astroman** » Thu Aug 19, 2004 5:08 pm

On 2004-08-19 06:07, janila wrote:
... The sample rate conversion from 96k to 44.1k is far more complex to do well than the conversion from 88.2k to 44.1k and the artifacts can be audible. ...

I'm not a math guru, so I cant't proove it, but it has been posted here (and elsewhere) that this is just a fairy tale

A proper samplerate conversion has the same complexity for any rate.

You probably refer to a 'fake' conversion where just every second sample is left out plus the neighbour values are interpolated.
That's a common trick (and more wide-spread than one might expect, specially in the reverse direction

), but no sample rate conversion - which is something with a 27th order filter math, bla, bla...

cheers, Tom