[sdiy] String filters and divide down oscillator string synths

[Tom Wiltshire]

Andrew,

On 21 Jun 2008, at 13:36, Andrew Simper wrote:

> I'm working on a digital emulation of a classic divide down string  
> synthesizers at the moment,


Sounds interesting. What are you using for that? Is this a VST or do  
you have particular hardware in mind?

> * 4 parallel band pass filters with adjustable freq, res, and gain
> * 1 high pass and low pass in series with freq, res and single gain  
> in parallel with a band pass filter with adjustable freq, res, and  
> gain, and then all that in series with a notch filter with freq and  
> res.
>
> Do you have any suggestions as to other possible designs? Do you  
> know of any links to any pages that have formant shapes for  
> instruments?

I'd probably try bandpass, like you suggest, implemented as state- 
variable filters.
I looked into formant shapes (a set of formant filters is definitely  
on my "one day..." list) but found precious little good information.  
It seems measuring the frequency responses of musical instruments  
isn't straightforward, and that few people have bothered.
Still, in the end I decided that all acoustic instruments have *some*  
resonances and dead spots and so forth, so having a section in your  
instrument that copies that effect is going to improve things, even  
if I don't know enough to realistically copy an existing instrument.
Looked at the other way around, one of the things that makes synths  
sound "Synthy" is the dead flat frequency response from 20-20KHz.

> I have done some research in the Arp Omini (I have one sitting in  
> front of me to test with) which others might be instereseted in  
> (sorry if this is all obvious stuff that everyone already knows)
>
> * To get a saw from a sqr divide down sqr wave you high pass filter  
> the sqr then clip off the bottom with a diode then high pass again.
> * The "amp" per note is achieved by varying a bias voltage to the  
> diode so the clip point changes. This results in some "thumping"  
> with note ons since the dc introduced has to be filtered away.
> * The "hollow" or waveform enhancement mode is made by moving the  
> bias voltage enough that the other spike of the high pass sqr  
> waveform is present in the output signal.
> * Since the amp per voice is made by varying the bias voltage for  
> the clip you get timbral variations through the attack and release.
> * The "string" sections simply put this waveform through a chorus.

This is another great example of the limitations of some very basic  
circuitry becoming the holy grail that more sophisticated circuitry  
strives to emulate (See also: Hammond organ, Minimoog filter,  
envelopes, Fender tube amplifiers). I always try and remember this  
when I'm designing something, and worrying too much about making the  
inevitable compromises. Instead, I should just get on and compromise!  
If it's successful, in twenty years time, someone will be using a 25  
Ziggahertz DSP to copy the exact quality of audio grunge my  
compromise caused! It's a funny old world!

Sorry, not a lot of useful information in that post - just some  
thoughts.

Regards,
Tom