Sound Morphing

[garyb]

going from a female voice to a male voice...sometimes the old-fashioned straight-forward way still works best.record both tracks and crossfade.no soft or hardware can do this any better.the hard and soft solutions are very cool.they just won't do that simple task better or easier.

[djody]

HE astro,

I Know my writting is not perfect but i'm trying to figure out what the spectral morphing principle is.

Ofcourse its also a parametric change in time, but the thing is that the modulation is not linear. It depends on the source material. each band has its own morph path. Wich is as dynamic as the sounds used.

Maybe you could rewrite my piece in proper english and with a nice build up

Laterz

[astroman]

djody, I think your writing describes the process very well, as far as that's possible at all with words
Without a listening example that's probably complete nonsense for someone who never heard about it.
We simply have different viewpoints: you go into the details and try to be as specific as possible, while I'm doing just the opposite and try to generalise as far as it gets
Nothing bad with that, it's possibly just my strange way of thinking influenced over the years of system design, support and troubleshooting which provide most of my income.

cheers, Tom

[kensuguro]

I think spectral morphing is the way to go. If you want to be able to choose between any sound. Drums->voice, male->female, donkeys->your boss, horns->strings. And everything inbetween. Only, since spectral morph uses either FFT STFT, for phase vocoding, you're likely to loose some attack. The drawback of any form of morphing, or in otherwords, mapping A's data onto B is that B always has to be wider in spectral content. Just like the old vocoder method, you don't get much by modulating a sin signal with your voice. The morphing voice example is always brought to attention because voices all generally have similar harmonic content and is easy to morph between.

STS5000's formant shifting algorythm is doing just the same. So, the code is all there. I'm just not sure why CW won't make a A>B morphing version of it. If they ever do release something, it'll probably be expensive, because it's a fairly known gimmick, and many people want it.

In the meanwhile, you can try it in:
windows:
csound, PD
mac:
supercollider(I think), Max/msp

Search for "phase vocoder" and you'll find tons of info. I believe Max/msp has one that does a good job of keeping phase information intact. (usual algos ruin the phase to some extent)

The problem with all FFT approaches is that it requires non-realtime analysis of the data.

A general way to do phase vocoding is to break down the audio into FFT, make alterations, and then do a reverse FFT resynthesis or additive resynthesis. Doing FFT on audio yields something like a graphic EQ, only you don't get a (freq,volume)plot from the equation. Internally, you get a bunch of strange numbers, in pairs of real and imaginary numbers.(strange?) FFT must have been thought up by a bunch of whacked up pot heads cuz it's really crazy.

<font size=-1>[ This Message was edited by: kensuguro on 2003-04-15 06:13 ]</font>