Hi!

As a spinoff of my efforts to create my very own, experimental audio codec, I have created a very nice audio effect. As an example, I processed the song "Radio #1" by the French band "Air". The processed sample is provided here (for educational purpose only. No copying!).

http://home.san.rr.com/sandiegodiaries/audio/chorale.mp3

I do not provide the original sample for copyright reasons. You will find it on filesharing systems if need be. In technical terms, what I am doing here can be described as follows:

1) Form FFTs of 65536 (2 ^ 16) samples, windowed (multiplied) with a sinusoidal function.

2) Transform real/imaginary coefficients into magnitude/phase representation

3) Randomize the phase

4) perform an inverse FFT

5) add the resulting signal to the output WAVE file and advance by 16384 samples (that is 1/4 of the transform size)

6) Loop till end of sample reached.

This can be seen as an extreme demonstration of the effect of "pre-echo" in audio coding. ;-) The effect of temporally smoothing/smearing transient signals is demonstrated here in a very exaggerated way.

The randomization of phase information removes any temporal localization of information within the transform block. Transient signals will be smeared over time. The sinusoidal window function blends the individual transform blocks so you don't hear any breaks or clicks between transform blocks.

You will notice that the percussion is smeared so badly that it almost sounds like breaking waves at the beach. And the vocals sound like a chorale in a cathedral, hence the name of the sample ;-)

Christian

Would this give you weird random surround effects as well? Cool

Do you have any code available for this?

I'd be interested in checking it out on a few different samples and see what happens.

Unfortunately, ... no.

I developed that analysis/synthesis code for an audiocodec.
This audio effect I tried was just a quick experiment. Meanwhile this program has evolved further:

Now it produces horribly big compressed audio files with bad artifacts. ;-)