Wavetables: Rude.

Wavetables: Rude.
« on: April 07, 2021, 12:04:46 PM »
I've been using Serum for 5+ years and have been experimenting with making my own wavetables for Pro3, but keep getting the same results

I've used the http://wav2pro3.beigemaze.com/ converters. Tried 2048, 1024..As well as converted some of my actual serum wavetables that I know sound great. - They just don't sound very good at all. [At least, not as good as they do in Serum]

Is this something I have to live with? Or are there any tricks.. Technically speaking, if you sample a, say, piano note and scan the wavetable correctly..It should sound like a near-real piano note. Not like robotic gobbelty-goop.

Pro3 OS: 1.1.2.8

Re: Wavetables: Rude.
« Reply #1 on: April 07, 2021, 12:41:03 PM »
Pro 3 has 16 keyframes, each at 1024 samples in length.   

Serum has up to 256 keyframes, and 2048 samples each. 

So, yeah, Serum has significantly higher resolution and capability to create more natural sounds with detail, or longer release trails.

I haven't tried piano samples, but I imagine if you just sample one note for a wavetable, split among the 16 keyframes,  you should be able to scan the wavetable with aux envelope and get a pretty decent piano sound, at least over an octave or two.     

As a side note / feature request, it would be really nice to have ability to switch at least the Aux Envelopes (and would be great on Filter/Amp too) to LINEAR contour ADR stages, for more accurate wavetable sample playback. 







OB-X8, Pro 3, P6, Rev2, Take 5, 3rd Wave, Deepmind, PolyBrute, Sub 37
Sound Sets:
https://sounddesign.sellfy.store/
Free Patches:
https://www.PresetPatch.com/user/CreativeSpiral

chysn

  • *****
  • 1812
Re: Wavetables: Rude.
« Reply #2 on: April 07, 2021, 01:17:49 PM »
Creativespiral said it really well.

WAV2Pro3 makes wavetable generation convenient by accepting 256-frame files directly from Serum. But the way it does this is by keeping every 16th frame and downsampling from 2048 samples to 1024. In other words, to meet the limitations of the Pro 3, WAV2Pro3 has to throw away almost 97% of the information it gets!

You can overcome this to a large extent by designing your wavetables in Serum specifically for the Pro 3. That is, create or add only 16 frames so that you have complete control over what gets into the Pro 3 wavetable.

Serum is the best tool available for creating Pro 3 wavetables, but using it for Pro 3 is a bit different than creating wavetables for use with the Serum synth itself.

I'm a bit skeptical about the prospect of scanning through a wavetable to approximate a 16384-sample cycle. For one thing, even though the Pro 3's frames are nominally 1024 samples, the 1024-sample set is only used at the very lowest part of the keyboard, up to around F#2. As you move up the keyboard, you're using 512, 256, and 128-sample versions of the waveform (Pro 3 wavetable sysex is so big because it really contains 64 waveforms). Also, scanning is going to be at a constant rate, not the variable rate that would be necessary to seamlessly link the waveforms together at different pitches. Plus you've got amplitude interpolation going on that you just can't turn off.

While some creative and dedicated programming might be able to get around some of these hurdles, my view is that it's probably fruitless to expect anything approaching "sample playback." Wavetable synthesis is a whole other thing, and it's hella fun when I'm not trying to make it jump through hoops. I treat sampled audio timbres more as raw material for synth sounds, with the bonus of motion.

And also, +1 for EG contour control.
« Last Edit: April 07, 2021, 02:15:33 PM by chysn »
Prophet 5 Rev 4 #2711

MPC One+ ∙ MuseScore 4

www.wav2pro3.comwww.soundcloud.com/beige-mazewww.github.com/chysnwww.beigemaze.com

he/him/his

Re: Wavetables: Rude.
« Reply #3 on: April 07, 2021, 03:34:04 PM »
Thank you for the great explanation!

Pro 3 has 16 keyframes, each at 1024 samples in length.   

Serum has up to 256 keyframes, and 2048 samples each. 

So, yeah, Serum has significantly higher resolution and capability to create more natural sounds with detail, or longer release trails.

I haven't tried piano samples, but I imagine if you just sample one note for a wavetable, split among the 16 keyframes,  you should be able to scan the wavetable with aux envelope and get a pretty decent piano sound, at least over an octave or two.     

As a side note / feature request, it would be really nice to have ability to switch at least the Aux Envelopes (and would be great on Filter/Amp too) to LINEAR contour ADR stages, for more accurate wavetable sample playback.
Pro3 OS: 1.1.2.8

Re: Wavetables: Rude.
« Reply #4 on: April 07, 2021, 03:34:21 PM »
Thank you!

Creativespiral said it really well.

WAV2Pro3 makes wavetable generation convenient by accepting 256-frame files directly from Serum. But the way it does this is by keeping every 16th frame and downsampling from 2048 samples to 1024. In other words, to meet the limitations of the Pro 3, WAV2Pro3 has to throw away almost 97% of the information it gets!

You can overcome this to a large extent by designing your wavetables in Serum specifically for the Pro 3. That is, create or add only 16 frames so that you have complete control over what gets into the Pro 3 wavetable.

Serum is the best tool available for creating Pro 3 wavetables, but using it for Pro 3 is a bit different than creating wavetables for use with the Serum synth itself.

I'm a bit skeptical about the prospect of scanning through a wavetable to approximate a 16384-sample cycle. For one thing, even though the Pro 3's frames are nominally 1024 samples, the 1024-sample set is only used at the very lowest part of the keyboard, up to around F#2. As you move up the keyboard, you're using 512, 256, and 128-sample versions of the waveform (Pro 3 wavetable sysex is so big because it really contains 64 waveforms). Also, scanning is going to be at a constant rate, not the variable rate that would be necessary to seamlessly link the waveforms together at different pitches. Plus you've got amplitude interpolation going on that you just can't turn off.

While some creative and dedicated programming might be able to get around some of these hurdles, my view is that it's probably fruitless to expect anything approaching "sample playback." Wavetable synthesis is a whole other thing, and it's hella fun when I'm not trying to make it jump through hoops. I treat sampled audio timbres more as raw material for synth sounds, with the bonus of motion.

And also, +1 for EG contour control.
Pro3 OS: 1.1.2.8

Re: Wavetables: Rude.
« Reply #5 on: April 07, 2021, 07:57:46 PM »
Plus you've got amplitude interpolation going on that you just can't turn off.

This I feel is the trickiest aspect to be aware of.  If your harmonics are out-of-phase between adjacent "keyframes", they drop out completely when interpolating between them.  Some of the built-in wavetables suffer from this, causing them to phase in and out while sweeping the waveshape parameter, rather than transitioning smoothly.

It is of course possible to automatically align harmonics between samples while maintaining zero crossings at the start and end (basically, cut out real components from the FFT), but, although this preserves the spectral content, when experimenting with this I noticed that it actually does change the sound.  This was surprising to me but apparently there are studies about this, see https://www.researchgate.net/publication/286939870_Sensitivity_of_Human_Hearing_to_Changes_in_Phase_Spectrum for example.

(The quality of this change in sound I find very peculiar.  E.g. randomizing the harmonic phases of a sawtooth wave creates a distinctive funky grungy distortion in the upper harmonics.  But strangest is that the sensation seems to fade from my auditory short-term memory very quickly, such that if I listen to a normal sawtooth immediately after a phase-randomized version, my brain doesn't register the sound as having changed, but then listening to the phase-randomized version again, the difference is readily apparent.)