getting ahead: convolution

[Martin Czech]

:::Danielson, not Danielsen. The later would be the Danish/Norwegian version of
:::the same name, but since I stubbornly consider myself a Swede and also was
:::given this name Danielson it stays!

Oh sorry! But still better the Bingo Debugs ;->


:::Now, one thing you still have not described is how the endresult sounds!
:::That is, have the methods you use know improved the audio quality compared to
:::the results you got with your older methods? I assume you had some interesting
:::gain in freedom and computational speed.


The improvements are on the computational side, and on the recording side.
I have tested a lot with the first version of the program and the methods
were not elaborated then, but the results were convincing.

The non deconvoluted room inpulse responses sounded very real,
but of course not as "beautiful" as artificial reverb. If you listen to  real
rooms you'll find a lot of "artefacts", like slapp echos and resonances, which
would lead to a very bad test result when coming out of a dsp reverb box.
Reality is sometimes not so nice! However, the reverbs don't have any 
repetitive character, as any IIR filter approach will finally have.
And simulations of shower cabinets or very small rooms are very
convincing, too.

And of course, how said that a reverb response has to be taken?
One could use a reverse reverb response (impossible with IIR, they do
it with gating), or a word or a sentence, or some noise, or ...

I certainly have not tryed all the interesting things.

But I did try "handmade" FIR filters, very impressive, too!
The good thing is that even very narrow bandwidth stuff comes out in a proper,
normalized fashion due to the 32bit intermediate format (I have never seen
a clipping up to now).

It would be nice to have some FIR-filter design program, be it optimal,
or windowing or sampling.


As I allready said: for the first time in my life the output was more
then the work that I have put into that program.
Usually intersting math things suck in the audio domain.

I still have to the the noise method for recording things in the studio.
I will burn an AUDIO-CD with such noise sequences, and record that
with 48kHz DAT. And of course I have to deconvolute the balloon impulse responses.

It is clear to me that I will gain a lot better frequency resonse, I've noticed
very early that the balloon stuff sounds hf damped and bass boosted.
So in a first approach I tryed to equalize that, the results where good.

Problem: Allmost no spare time in the moment, as I do >600 km cycling per week.
If you don't slim in the winter, you have to do it in the hard way in spring.
After a >200km practising unit you don't feel like typing some more C-code.

Ok, this work is now going on since Oct 99, so there is no reason why I should drop
it, it will just take some time.

m.c.