Sound Synthesis

[Tony Hardie-Bick]

Tom Davies wrote:
> Subtractive, gotta be eh
> 
>> What is the best type of synthesis for a beginner to start with?
>>
>> FM? Sample? Wavetable? Subtractive? Additive? Physical Modelling?
>> Granular?  
>>
>> Something else?

FM is by far the easiest. I say this, with my own particular bias 
against the negative aspects of digital synthesis (ie, all the stuff 
that causes people to think digital is inferior to analogue). With FM, 
you have a sine oscillator (use a lookup table of 1024 points with 
linear interpolation, or 65K points without). Equip yourself with 
"Musical Applications of Microprocessors" by Hal Chamberlin, and just 
use the oscilltor output as a phase modulation source to other sine 
oscs, and/or to itself.

If you go the subtractive route you're not gonna be ready to implement 
an alias free square or sawtooth oscillator. That stuff is 
counter-intuitive, and not fun until you've really hit on lots of other 
tricks first. Heavy on the maths. Tough on the codability too.

Plenty of books will tell you to make a sawtooth waveform by simply 
adding a number to a register (or integer variable) and let it 
wraparound. Well, fine. That will give you aliasing. Same if you use a 
conditional to get square and pulse width, unless you use a really high 
sample rate (8x or 16x is beginning to be okay). The whole reason there 
are still analogue synths in production, is because digital synths got a 
bad name from these techniques. Personally, I believe digital is a 
fantastic unlimited resource for acoustic aesthetics of unlimited 
beauty, and have refused to compromise on such issues - so, up till now, 
I have used FM in all the digital synths I built and programmed.

Filters are another area where digital has given itself a bit of a bad 
name. Depends on what you're after. Many digital synths have simple, 
well-implemented, high quality non-distorting filters. These are not 
"warm" or crazy, or particularly characterful, but in the cold aesthetic 
of the digital synth that somewhat prevails, this may in fact be an 
enormous advantage (many excellent synth products fall in this category, 
incl. all Virus models, Korg MS2000, all Roland's digital stuff). There 
are only a couple of digital synths that seem to have captured some of 
the roughness of analogue filters, and these are: Korg Legacy series 
MS20, Elektron synths with SID emulation. Yamaha AN1X gets an honourable 
mention. Many synths place a distortion stage (sigmoid LUT) after the 
filter, which is the surest way to make a dull filter sound nasty, as 
well as dull. The distortion belongs inside the iterated signal feedback 
paths (ie, applied to a state variable) INSIDE the filter, not outside. 
You need to choose your algorithm so that the overdrive is fed with an 
appropriate signal. If you want to look at a no-compromise filter 
written in C, you can grab this:

http://www.entity.net/DFM-1/downloads/dfm1.tar.gz

musicdsp.org has a load of algorithms, but you'd need to read a lot 
before you'd get the big picture, and still some info seems missing from 
there.

One great thing about FM, is you get variable spectra, without having to 
do any filtering, and you can add that when you want to get into it.

Digital without cheese costs extra. Depends on what you're looking for. 
Personally, I love analogue sounds, but would like to know *why* that 
stuff sounds good, and put it down in code so I, and anybody else can 
use it. FM is a great way to pursue this goal keeping the code intuitive 
and therefore inline with one's goal of exploring the aesthetics of 
musical programming.

Oh. Did I write all that?

Tony (HB)