Help with wavetable synthesis

[dayunbao]

I haven't worked with wavetable synthesis much before, but decided to start playing around with it. So I watched Eli Fieldsteel's 2 videos, and have been trying to reproduce some of the things he's done. This link shows roughly what Eli is doing that I'd like to reproduce. However, I'm having some issues translating that to Supriya. For example, I can't use ExpRand in the same way he did, since I believe I need to create the buffers outside of the SynthDef. That means I can't use the ExpRand UGen because it returns an OutputProxy not a float, and I'm not sure what's the best equivalent Python function. I've been trying random.uniform(), but I think that's causing the waveforms to all look too similar (see plots), unless there's something going on with the plotting function. I have noticed that the wave forms all have an offset of some kind. I'm not sure why that is. Another thing I'm not quite sure how to reproduce is this: LFNoise1.kr(frequency=0.5).range(0, num_buffs). My solution was to try to use LinLin instead of .range(...), but I'm not sure if I'm mapping it correctly.

Here's some test code:

import random
import sys

import supriya
from supriya import Buffer, Envelope, Server, synthdef
from supriya.ugens import (
    Envelope,
    LinLin, 
    Out, 
)
from supriya.ugens.filters import LeakDC
from supriya.ugens.mac import MouseX
from supriya.ugens.noise import  LFNoise1
from supriya.ugens.osc import Osc, VOsc

@synthdef()
def variable_wavetable(buf_start_num, num_buffs, frequency=440.0, phase=0.0):
    noise = LFNoise1.kr(frequency=0.5)
    buf_pos = LinLin.kr(
        source=noise,
        input_minimum=0.0,
        input_maximum=1.0,
        output_minimum=buf_start_num,
        output_maximum=num_buffs - 1,
    )
    signal = VOsc.ar(
        buffer_id=buf_pos, 
        frequency=frequency, 
        phase=phase
    )
    signal = LeakDC.ar(source=signal)
    signal *= 0.1
    Out.ar(bus=0, source=[signal, signal])

def create_wavetable_buffer(server) :
    buffer = server.add_buffer(channel_count=1, frame_count=2048)
    num_segments = random.randrange(4, 20)
    
    amplitudes = []
    for _ in range(num_segments + 1):
        amplitudes.append(random.uniform(-1.0, 1.0))
    
    durations = []
    curves = []
    for _ in range(num_segments):
        durations.append(random.uniform(1.0, 20.0))
        curves.append(random.uniform(-20.0, 20.0))

    amps = Envelope(
        amplitudes=amplitudes,
        durations=durations,
        curves=curves,
    )
    
    buffer.generate(
        command_name='sine1',
        amplitudes=amps.serialize(),
        should_normalize=True,
        as_wavetable=True,
    )

def main() -> None:
    server = Server().boot()
    server.add_synthdefs(variable_wavetable)
    server.sync()
    
    create_wavetable_buffer(server=server)
    create_wavetable_buffer(server=server)
    create_wavetable_buffer(server=server)
    create_wavetable_buffer(server=server)

    server.add_synth(
        buf_start_num=0, 
        frequency=120.0,
        num_buffs=4, 
        synthdef=variable_wavetable,
    )

    while True:
        continue

if __name__ == '__main__':
    try:
        main()
    except KeyboardInterrupt:
        sys.exit(0)

Plotted waveforms:

[josephine-wolf-oberholtzer]

1. sclang's range() method knows about unipolar vs bipolar UGens, but supriya's scale() method doesn't and requires you manually punch in the initial range (e.g. 0 to 1 vs -1 to 1).

2. Are you sure you want amps.serialize()? That's spitting out the envelope format required for use with an EnvGen UGen. It will look pretty random if you use it as a series of sine1 amps, but maybe that's what you wanted?

[josephine-wolf-oberholtzer]

Nobody ever asked about this before, and I never tried it myself, so it probably has to be implemented from scratch.

[dayunbao]

Nobody ever asked about this before, and I never tried it myself, so it probably has to be implemented from scratch.

OK.

For the record, I am trying to make random sine1s by generating random values for amplitudes, durations, and curves. I also was able to get waveforms that pretty closely matched what Eli got when not using random values for those parameters.

[dayunbao]

Do you know of a good equivalent to SuperCollider's/Supriya's ExpRand Ugen in normal Python?

[josephine-wolf-oberholtzer]

As part of a SynthDef or just as a language-side random number generator?

[josephine-wolf-oberholtzer]

Maybe you want https://docs.python.org/3/library/random.html#random.expovariate ?

[dayunbao]

As part of a language-side random number generator. I've looked at expovariate, but it doesn't accept a range. I could scale it after the fact, I guess...

…

On Sun, Apr 20, 2025, 4:05 PM Joséphine Wolf Oberholtzer < ***@***.***> wrote: Maybe you want https://docs.python.org/3/library/random.html#random.expovariate ? — Reply to this email directly, view it on GitHub <#492 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVNNEWODFGSZCMBMTEOAYD22OSQ5AVCNFSM6AAAAAB3PVEWAGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEOBZGA3DINQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[josephine-wolf-oberholtzer]

Added an issue to implement envelope-to-float-array logic: #493

I'm unlikely to get to this since I'm busy with some very involved new feature development, but if you wanted to contribute back, reverse-engineering sclang's logic here would be great.

[dayunbao]

Cool! I'll try to take a look later. I'm having a surgery tomorrow, and so will be out of action for a while. I've been told the recovery will be quick, but the results of the planned biopsy will dictate what's going to happen in the coming months. FM synthesis, wavetable synthesis, and granular synthesis are three areas I don't know much about, and want to better understand. Today I started looking in FM synthesis. I'll probably play around with that for a while because I don't need any special functionality to do it, unlike wavetable synthesis. I've become quite interested in modulation, in general, since I've been playing around with VCV Rack. It's got me thinking about how to incorporate more sophisticated modulation in my project. I was originally planning to explore that with wavetable synthesis, but FM synthesis is just as good.

…

On Mon, Apr 21, 2025, 3:50 PM Joséphine Wolf Oberholtzer < ***@***.***> wrote: Added an issue to implement envelope-to-float-array logic: #493 <#493> I'm unlikely to get to this since I'm busy with some very involved new feature development, but if you wanted to contribute back, reverse-engineering sclang's logic here would be great. — Reply to this email directly, view it on GitHub <#492 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVNNEQVYXYFF5CZAFS2U3T22TZTDAVCNFSM6AAAAAB3PVEWAGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEOBZHA4TOMQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[josephine-wolf-oberholtzer]

Oh, take care then. Zero pressure. I know just the sedation alone takes a lot out of me.