Hello,

I've got a granular synthesis patch that I've tried to build from the ground up. I don't really know all that much about granular synthesis, and I'm having issues with phasing.

If you record something into the patch (by clicking the record box and talking into your mic for 7 seconds, then unclicking the record box), then try and play things back, you'll notice some strange things, especially as you change the "rate" slider. If you record yourself humming one pitch for 7 seconds and then move the "rate" slider around, it seems that the sound goes in and out of phase, hitting several sweet spots along the slider's width. I know this must have to do with the fact that the same little snippet of the table is being played by enveloped, slightly overlapping tabread4's (in the granplay~ abstraction, in the "pd granplayers" subpatch), so sometimes adjacent periods of the wave in the table cancel each other out, and sometimes they reinforce each other. I also know that the number of sweet spots is related to the number of granplay~ objects I use (would these be called "voices?").

If I get rid of three of the granplay~ objects and just keep two of them, setting their offsets (argument $2) to 0 and .5, respectively, the problem mostly goes away, except the result is a lot choppier.

I want a really, really smooth way of doing what I'm trying to do here. I'm thinking that maybe I should go the fft route, unless there's a simple way to fix this patch I've got.

Any advice / help?

Thanks!

http://www.pdpatchrepo.info/hurleur/jake2_granular.zip

phasing is a classic problem faced by granular synthesis.

the general way to reduce it is to add some 'jitter' to the various playback parameters. what this means is that for example, with your offsets, you have them set to 0 and 0.5. but what you can do, is add a little random amount to the offset every time the grain is triggered.

subtlety is the key here though. you want to add just enough to offset the phasing effects, but not too much to radically/chop up alter your sound.

Or you can try pitch synchronous granular synthesis. It's very similar to overlap add pitch shifting methods. If the material is at F then only pick grains at n(1/F) intervals from a n(1/F) time grid. It works okay over short windows with strongly pitched instruments like strings and brass, but it breaks down with longer files and you're back where you started with phasing problems.

Thanks to both of you. I'll eventually try both.

Hardoff, the grains are never triggered in my patch - I have them being driven by a phasor whose phase is changed for each granplay~ and sent to a tabread4~, whose output is enveloped by the positive part of a cos~. Do you think it might make sense to just quietly change the offset during the part of the cycle when the envelope is at zero? That way, would I avoid having to actually trigger things and get around the delay timing inaccuracies I've been reading about? Or would I get a bunch of clicking because a delay would not allow me to change the offset at the precise moment when the envelope is at 0?

Also, is there a better envelope to use than a cos~? I was thinking maybe something like -(x^6)+1 would be closer to an "equal power" fade - is that correct?

Thanks so so much.

If you need variable slope envelopes you might want to look at this granulator --

https://puredata.info/Members/davigoli

s

I don't think I need variable slope envelopes - I'm just trying to put something better in place of cos~.

you could try

[phasor~]
|
[cos~]
|
[-~ 0.5]
|
[*~ 0.5]
|
[clip~ -0.1 0.1]
|
[*~ 10]

for a more rasied cosine shape?

If i'm not helping tell me to shut up

>>Do you think it might make sense to just quietly change the offset during the part of the cycle when the envelope is at zero? <<

yes, totally. i think you should do that.

of course, i would recommend to quit using phasor~ and start using vline~ instead, but it's probably not too much of a problem.

one thing about phasor~ is that its phase will only be reset once every block (default 64 samples = 1.5ms @ 44100hz samplerate)

so, if you cut your volume to 0, make sure it stays at zero for longer than 64 samples, or otherwise you will get cuts.

no...

the more i think about it, the more i think you should do what everyone else does and use vline~ 's to trigger your sample playback and also envelopes.

I seem to have it working fine like this. The nice thing about the phasor is that I don't really have to specify for each instance/voice when to trigger - each one can just add its offset to the main phasor as specified in the creation argument. Makes it really easy to add or remove instances.

Would it significantly reduce processor load to use vline~ instead?

hey myoosik

this granular patch is awesome! i'm looking for a way to build my own, but i'm a beginner at PD.

i was wondering if anyone had the time to maybe add some comments into this patch for me to download and observe

it think it would really help me on my way!
.dataplex