[sdiy] Random DSP thoughts

[Richard Wentk]

On 26 Jan 2009, at 19:42, Aaron Lanterman wrote:

>
> On Jan 26, 2009, at 2:37 PM, Michael Bechard wrote:
>
>>> Inharmonic Oscillator. Venture into pseudo-additive synthesis  
>>> with an
>> oscillator whose harmonics are variable in frequency. Be able to have
>> the harmonics shift over the note duration.
>
> I'll throw in that this was one of the main strengths of the  
> Digital Keyboards Synergy. Each partial had its own amplitude AND  
> pitch envelope, and these envelopes could be quite complex.. Wendy  
> Carlos talks about this on "The Secrets of Synthesis" if I recall  
> correctly.

You can do this kind of thing in real time with Csound or Max/MSP -  
do real-time FFT analysis of a sound file, use the coefficients to  
drive an oscillator bank or inverse FFT with some massaging of the data.

Massaging can mean pitch quantisation to specific harmonic grids (not  
just 1,2,3... but chord patterns), stretching, squashing, linear  
shifting, elimination for filtering, spectral multiplication with a  
different source - and so on.

Or you can generate the coefficients numerically/statistically.

It's actually hard to get right. The oscillator bank approach isn't  
so bad, but if you do inverse FFTs you need a lot of overlapping and  
smoothing for a non-lumpy result, which usually maxes out even fast  
processors, especially if you try to do it polyphonically.

Cameleon's Camel 5000 VSTi does some of this in real time. A couple  
of other VSTi's do variations on the theme.

The Synergy stuff was kind of limited, in practice. For good  
synthesis you need a 3D table of coefficients which includes  
interpolation of partials taking into account both pitch and velocity  
changes.

There's no point synthesizing a violin spectrum at a single frequency  
- the real thing includes fixed resonant modes, which a single sample  
won't model properly. But you can get some goofy digital sounds with  
wandering partials which are hard to create any other way.

Richard