Forums

Diffusion using convolution

Started by Zot March 12, 2007
Hi folks, I was experimenting with convolution, using the the overlap-add 
ideas with FFT. I was using an impulse of  fading white noise to test the 
algorithm for smoothness, and I noticed to made an almost perfect diffusion 
sound.

My question; is it possible to feed the FFT with white noise and a mixture 
of the previous noise or something to create a nice diffusion fall-off 
without using a huge noise table? Basically having a nice long diffusion 
tail without the massive impulse. I have tried diminishing the noise then 
adding a new random set on top with each frame, but I only got very 
repetitive buzzing, could this or something similar actually work?

Thanks,
Dave Hoskins



Oh, nevermind. Be all that ends all.   : ) If you know anyone that says the 
words "Paradigm shift,"  then your with the wrong company.  : ) ...... ..... 
.... ... .. .




"Zot" <bob@yahoo.com> wrote in message 
news:J2cJh.132269$HO5.95382@fe1.news.blueyonder.co.uk...
> Hi folks, I was experimenting with convolution, using the the overlap-add > ideas with FFT. I was using an impulse of fading white noise to test the > algorithm for smoothness, and I noticed to made an almost perfect > diffusion sound. > > My question; is it possible to feed the FFT with white noise and a mixture > of the previous noise or something to create a nice diffusion fall-off > without using a huge noise table? Basically having a nice long diffusion > tail without the massive impulse. I have tried diminishing the noise then > adding a new random set on top with each frame, but I only got very > repetitive buzzing, could this or something similar actually work? > > Thanks, > Dave Hoskins > > >
What is a "diffusion sound" ?

On 13 Mrz., 05:19, "Zot" <b...@yahoo.com> wrote:
> Oh, nevermind. Be all that ends all. : ) If you know anyone that says the > words "Paradigm shift," then your with the wrong company. : ) ...... ..... > .... ... .. . > > "Zot" <b...@yahoo.com> wrote in message > > news:J2cJh.132269$HO5.95382@fe1.news.blueyonder.co.uk... > > > > > Hi folks, I was experimenting with convolution, using the the overlap-add > > ideas with FFT. I was using an impulse of fading white noise to test the > > algorithm for smoothness, and I noticed to made an almost perfect > > diffusion sound. > > > My question; is it possible to feed the FFT with white noise and a mixture > > of the previous noise or something to create a nice diffusion fall-off > > without using a huge noise table? Basically having a nice long diffusion > > tail without the massive impulse. I have tried diminishing the noise then > > adding a new random set on top with each frame, but I only got very > > repetitive buzzing, could this or something similar actually work? > > > Thanks, > > Dave Hoskins- Zitierten Text ausblenden - > > - Zitierten Text anzeigen -
In diffusion, I meant something similar to the long drawn out tail of a 
reverb unit. Normally done with a bank of different length delay lines in 
parallel to blur the sound. The trouble with this method is that you have to 
apply a lot of delays and add modulation and countless listening tests and 
tweaking to make it acceptable and non-metallic - Which is fine, but I 
noticed that convolving a white noise signal with it did a fine job also. So 
I was wondering if it was a valid solution, and also if the impulse could be 
adjusted each frame instead of having a giant fading white noise impulse for 
long tails.
I hope that makes sense!

Dave



"Andor" <andor.bariska@gmail.com> wrote in message 
news:1173888952.923873.302240@n76g2000hsh.googlegroups.com...
> What is a "diffusion sound" ? > > On 13 Mrz., 05:19, "Zot" <b...@yahoo.com> wrote: >> Oh, nevermind. Be all that ends all. >> >> "Zot" <b...@yahoo.com> wrote in message >> >> news:J2cJh.132269$HO5.95382@fe1.news.blueyonder.co.uk... >> >> >> >> > Hi folks, I was experimenting with convolution, using the the >> > overlap-add >> > ideas with FFT. I was using an impulse of fading white noise to test >> > the >> > algorithm for smoothness, and I noticed to made an almost perfect >> > diffusion sound. >> >> > My question; is it possible to feed the FFT with white noise and a >> > mixture >> > of the previous noise or something to create a nice diffusion fall-off >> > without using a huge noise table? Basically having a nice long >> > diffusion >> > tail without the massive impulse. I have tried diminishing the noise >> > then >> > adding a new random set on top with each frame, but I only got very >> > repetitive buzzing, could this or something similar actually work? >> >> > Thanks, >> > Dave Hoskins- Zitierten Text ausblenden - >> >> - Zitierten Text anzeigen - > >
Andor wrote:

   ...

>> - Zitierten Text anzeigen -
Zitierten Text anzeigen? Verwenden Sie Google? Jerry -- Engineering is the art of making what you want from things you can get. &macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;
On 14 Mrz., 23:10, Jerry Avins <j...@ieee.org> wrote:
> Andor wrote: > > ... > > >> - Zitierten Text anzeigen - > > Zitierten Text anzeigen? > > Verwenden Sie Google?
Aber klar doch! Gruss, Andor
I guess I'm either barking up the wrong tree or exactly the right one!

Nizza, wenn jemand mir einen Tip gab.

"Andor" <andor.bariska@gmail.com> wrote in message 
news:1173946784.480496.321400@y66g2000hsf.googlegroups.com...
> On 14 Mrz., 23:10, Jerry Avins <j...@ieee.org> wrote: >> Andor wrote: >> >> ... >> >> >> - Zitierten Text anzeigen - >> >> Zitierten Text anzeigen? >> >> Verwenden Sie Google? > > Aber klar doch! > > Gruss, > Andor >

Zot wrote:
> I guess I'm either barking up the wrong tree or exactly the right one!
I still don't know exactly what your question is. You want to use white noise as reverb, and somehow compress the white noise to reduce the "table size" (what table?)? You know that you can't compress white noise (by definition). Can you elaborate on your goal / problem a bit?
O.K. thanks.
If I convolve a sample of say a drum with a long fading out white noise 
impulse, I get a really nice gradual fading drum sound, that's all and mixed 
up and blurred out through time, which is what I termed diffusion.
If I'm doing a 1024 point FFT to convolve through this white noise, do I 
really need the large white noise impulse, because large convolutions take 
longer to calc? Is there a way to populate the 1024 FFT every block using 
some kind of residue of the old FFT and  new random numbers?  The more I 
(try to) explain this, the more it sounds unlikely, as each sample has to be 
multiplied by the whole of the existing noise sample.. hmmm... head hurting, 
must stop!

Cheers,

Dave


"Andor" <andor.bariska@gmail.com> wrote in message 
news:1174069061.703709.286660@n59g2000hsh.googlegroups.com...
> > > Zot wrote: >> I guess I'm either barking up the wrong tree or exactly the right one! > > I still don't know exactly what your question is. You want to use > white noise as reverb, and somehow compress the white noise to reduce > the "table size" (what table?)? You know that you can't compress white > noise (by definition). Can you elaborate on your goal / problem a bit? >
I think what you are looking for is how to approximate a given impulse
response (FIR system) through a system with rational transfer
function. You can start searching using "model order reduction" and
"system identification".


Zot wrote:
> O.K. thanks. > If I convolve a sample of say a drum with a long fading out white noise > impulse, I get a really nice gradual fading drum sound, that's all and mixed > up and blurred out through time, which is what I termed diffusion. > If I'm doing a 1024 point FFT to convolve through this white noise, do I > really need the large white noise impulse, because large convolutions take > longer to calc? Is there a way to populate the 1024 FFT every block using > some kind of residue of the old FFT and new random numbers? The more I > (try to) explain this, the more it sounds unlikely, as each sample has to be > multiplied by the whole of the existing noise sample.. hmmm... head hurting, > must stop! > > Cheers, > > Dave > > "Andor" <andor.bari...@gmail.com> wrote in message > > news:1174069061.703709.286660@n59g2000hsh.googlegroups.com... > > > > > > > Zot wrote: > >> I guess I'm either barking up the wrong tree or exactly the right one! > > > I still don't know exactly what your question is. You want to use > > white noise as reverb, and somehow compress the white noise to reduce > > the "table size" (what table?)? You know that you can't compress white > > noise (by definition). Can you elaborate on your goal / problem a bit?- Zitierten Text ausblenden - > > - Zitierten Text anzeigen -