@Another User said:

We have farms of PCs for samples, why not have farms of PCs for the virtual space also?

All the best,
PolarBear

This would make things even more complicated, as it seems.

@SyQuEsT said:

Why not a DSP plugin (up to 9 processors !) for SSL-Sydec Soundscape (www.sydec.be) and/or TC-6000 / TC-Icon ...

Because convolution needs much, much more memory than all these DSP-"farms" offer (apart form the "political" reasons) -- remember: we are talking about _sampling_reverb. All the sampled IR-data has to be available for the CPU in real-time.

Lake Technologies once had the Huron DSP board for complex convolution processing. It seems to have diappeared, hasn't it?

Yeah, I know about the problems with DSP and RAM if you want to do a REAL convolution processing (and not that fake one which is in all these plugins). I urged Symbolic Sound to implement convolution in their Kyma Workstation for a long time but even for a full blown system you just could convolute with a 0.8 seconds impulse response. There a technologies for bringing the dsp load down, but they are patentet by Yamaha and Sony, if I'm not totally off.
So, there we go...

@mathis said:

[...] I know about the problems with DSP and RAM if you want to do a REAL convolution processing (and not that fake one which is in all these plugins). [...]

I have to admit that I don't know what you mean by this ...

In pure time convolution - as far as I understood it which is not said that I really understood it - you have to multiply every single sample with every single sample of the IR which consumes enormous dsp power.
There are methods using a combination of time and frequency convolution which help the multiplying engine by using some FFT processing. I was told that all software plugins use this method to be able to do convolution at all. But that's not the real thing like it is applied in the Sony and Yamaha standalone units (are they still built, b.t.w.?) They use obviously other methods to bring dsp load down. Symbolic Sound didn't want to apply any other than pure time convolution and couldn't because of patents.

Probably it's not all right in the details what I wrote here, but the general outline should be true.

Now I understand. For what we are after, it's the result that counts, so every way to make the way to this result faster and easier is a valid one. You know - if it sounds right, it is right [[;)]]

Hmmm, I only recently read some interesting things about this topic on another topic. Confusing? Not so: wave field synthesis uses convolution to simulate real spaces through arrays of speakers surrounding the listener. There are installments for research in a small cinema in Germany, Ilmenau, at the Fraunhofer Institut, as well as it was already used at the Bregenzer Seebühne. Still a work in progress as I see so far, but it's said it's only possible to do nowadays since its discovery, because - and that's the interesting part - the computing power is availible today... from 44 up to 192 speakers have to be fed with signals and every "source" in the soudn image needs one convolution. They say, they only use the recorded impulse for a certain, more or less short, time and simulate the rest of the impulse with a conventional reverb, in order not to save some substantial computing power.

I'm sure you already know about these technologies, but maybe the principles of these could also be applied to your technology here? I don't know...

All the best,
PolarBear

@PolarBear said:

[...] in Germany, Ilmenau, at the Fraunhofer Institut [...]
PolarBear

... now guess where I go to in December (not for the first time, BTW) [;)]

@Dietz said:

Now I understand. For what we are after, it's the result that counts, so every way to make the way to this result faster and easier is a valid one. You know - if it sounds right, it is right [[;)]]

Is the result actually any different? I never heard a comparison of the varying convolution techniques. There should be a difference, right?

Theoretically the results are identical. There are certain caveats and restrictions in real life, but their effects are marginal compared to the benefits.

I'm trying to understand what you mean by frequency convolution. Do you mean constructing filters by using a static "picture?" Time-compressing sweeps?

What I don't understand is how you get the decay time of whatever you're sampling that way - assuming I'm on the right track.

(Before a little over a year ago I didn't understand pure time convolution either, so please be patient. [:)] )

http://www.sfu.ca/~truax/conv.html

This explains it, but it's still not very clear in my brain.

Finally we still don't know when it will be available. I do need it before the end of January, do you think I still have some chance?

No, definitely not.

DAMNED !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Ok...now I'm depressed!!!!.....

when do you think it would be the sooner?

Please understand that I can't give you more detailed information.

Iunderstood it was that CPUs are just not efficient enough yet.

I assume the recent Core 2 Duos are not up to it as they've been out for a while. Perhaps the "Core 2 Extreme Quad" processors will be enough. They're due out in November and are supposedly 70% faster than the Core 2 Duos.

And by the end of the decade processors with 80 cores and teraflop performance on a single PC and essentially unlimited RAM.

We must be patient - and I suspect it's more frustrating for Dietz than anyone else.

Hey, just get yourself one of these, Dietz!

http://www.channelregister.co.uk/2006/11/06/japan_512-core_co-pro/

that should do it... [;)]

J.