Objective

To explore using convolution to create new sounds.

Strategy

There are six convolution opcodes in csound; I'm going to test each one out. My idea here is to convolve with 1.) something normal, 2.) something interesting, and 3.) something bizarre. I'm starting with convolve, which, according to the documentation, entered the opcode world in 1996.


; Csound Convolution Opcodes
; ==========================
; convolve 1996
; cross2 1997
; dconv 2001
; ftmorf 2002
; pconvolve 2004
; ftconv 2005


Materials

I'm using five different sounds:

• FluteA4.wav, OboeA4.wav, and ViolinA4.wav -- Cut out of samples from the University of Iowa's
  Musical Instrument Samples
• RideCymbal.wav -- from ltibbits on Freesound.
• OmniCenter25ft.wav -- from Noisevault

Only the last one is an actual 'reverb' type sound file.

Result

OK, when you use cvanal on even a small sound file, it turns into a monster truck sized convolution file. Therefore, for the first time since my return, the .csd files below will actually not run on your machine -- I didn't include the sound files OR the convolution files (I figured yahoos like me might eat up all the server space) in the .csd structure.

However, I did post the results. You can read the csound files to see what I did -- or you can listen to the .ogg files and hear the result!

cvOmni.ogg is the 'normal' file -- the sounds of a cymbal, flute, oboe, and violin in a really large hall (convolved with 'OmniCenter25ft.cv').

cvRide.ogg is the 'interesting' file -- and the one I like the most. I convolved here with the ride cymbal. Nice! What an eerie sound this gives each instrument.

cvOboe.ogg is the 'bizarre' file -- I convolved with the Oboe, which is about as far away from an impulse as you can get. At least it's one of the shorter samples ... I know, mathematically, things that don't decay tend to ramp up to huge amplitudes rather quickly. Here I had to do a lot of manual scaling to get the sounds to stay in the 16 bit limit.

Conclusion

This was one of the best experiments to date. Everything worked correctly the first time. Makes sense -- if the opcode is from 1996, and hasn't been replaced by something better, it must be pretty good. Certainly there are five more options, but each seems to be a different spin on the same idea.

But we'll see.