Re: [Audacity-nyquist] Here's a phase distortion example

[Roger D. <rb...@cs...>]

If I'm understanding the example, on each period, you want to traverse 
half wave table linearly within some fraction dist of the total period, 
and then traverse the rest of the table linearly for the remainder of 
the period (please correct me if I'm wrong).

This is a special case of FM: you are modulating the frequency up and 
down with a rectangular wave. Ignoring rounding error, you could do this 
with Nyquist's FMOSC, but then you are left with the problem of creating 
the modulator and I suspect that the phase would drift due to rounding 
error whereas it seems like this algorithm assumes the modulator is 
synchronized to waveform traversal (the up/down frequency modulation 
occurs at the same phase every period).

Another possibility is to use SND-COMPOSE, which computes the functional 
composition of two signals: f(g(t)). You'd want to make f() be just the 
output of an oscillator with a nice integer number of samples per 
period. Then g() is the modulator: a monotonically increasing piece-wise 
linear function. You'd have to construct this by splicing together 
linear segments, probably using  something like (PWLV start dur end) 
inside a seqrep().

SND-COMPOSE uses linear interpolation, so the quality won't be really 
high, but then the modulator is not band-limited so there will be 
aliasing in any case. At least if you're careful you can get a subsample 
accurate g() which will help make slow modulation of the distortion 
parameter result in a slowly changing waveform as opposed to one where 
changes are quantized to sample boundaries.

This is definitely the high-level view. Let me know if you need more 
detailed help.

Also, in the example code, the use of SEQ is not correct (I can see how 
it might work due to some implementation short-cuts in Nyquist). The 
main problem is this will build a huge tree of unevaluated/suspended 
SOUNDS, probably a couple of kbytes per period.

The "right" way to do this would look something like
(seqrep (i n) (let (...) ... (seq (compute-a) (compute-b))))

or in SAL, it's more like

begin
    with ... initialize variables ...
    return seqrep(i, n, seq(compute-a(), compute-b()))
end

-Roger