I have a patch with 80 bangs that reference the same abstraction. When I click one of the bangs, it simultaneously calls the abstraction three times to produce a chord with three notes. Most of the time it sounds correct, but I often get a crackling noise. The noise usually starts at the middle or end of the chord. I've tried changing all of the audio parameters. Most of the parameters did not affect the noise. Changing the sample rate to 96k made the problem worse. Increasing the block size to 2048 made the problem better, but did not eliminate it. How can I get rid of the noise?

I'm running on Windows 10. I've tried both the internal sound card and an external USB sound card (Sound Blaster Play! 3). I've also tried the patch on a different Windows PC. The problem occurs on every PC and sound card.

@jamcultur Would you create a zip folder containing all the parts of the patch please and upload it to the forum.
David.

Here's the zip file:
tritone.zip

I don't know whether it's related, but I couldn't try one of the PD audio options. I'm using "old MMIO system". If I try to change to "standard (portaudio)", I get the message, "error opening audio: Invalid sample rate", no matter what the sample rate is set to.

One correction to what I originally posted... there aren't 80 bangs, there are 81 groups of 3 bangs, so 243 bangs altogether that can each invoke the abstraction 3 times.

@jamcultur Well. It takes a while to load and will only produce distorted notes for me on an old MacBook Pro I7 running windows 7.
I tried changing [vline~] to [line] as you have 729 of them and [vline~] is quite cpu hungry.
That reduced processor load by about 50% so it might work for you.
You can keep the same messages, but might need to set a smaller "grain" for [line] than its default.

It is then still bad on my system, but there are 4374 [osc~] objects loaded, and that is rather a lot.
Even moving the mouse across my laptop screen halts the audio.

Do you need all of them at once or could you set up "voice stealing" like in the [poly] object.
I.E. How many chords will be played at once?
Looking at the main patch in a text editor it is obvious that you could also have an abstraction for the 81 voice groups..... setting the pitches with arguments ($1 $2... etc.).... although if you opt for voice stealing you would simply send a list of the notes into one of them.
[clone] you might find useful as a container for the abstractions.
See this thread..... https://forum.pdpatchrepo.info/topic/9774/pure-data-noob but especially the part "all about dollars"

As a general rule, when you find yourself copying and pasting part of your patch then it is time to use an abstraction...!
David.

I changed [vline~] to [line] as you suggested, but that introduced a new noise. [line~] worked better.

I'll try making an abstraction for the 81 groups. That will reduce the number of [osc~] objects to 1458. Do you think that will be enough to eliminate the noise?

I typically only play one chord at time, but I was hoping to be able to play two or three. I don't know how to implement voice stealing, but I'm starting to research it. Do you know of any examples of voice stealing I could look at?

Thanks for your help.

@jamcultur You're very welcome.
There is a long recent thread on "poly" here....... https://forum.pdpatchrepo.info/topic/14537/poly-for-anything although the OP wanted to sort data by a leading symbol.
At the end it was realised that a good solution would be to get the voice itself let the patch know when it was finished and could be re-used.
There are a few good methods tried in the thread, with patches attached.
One of them might work straight away for you.
David.

I reduced the number of [osc~] objects to 27, and still have noise. The noise doesn't sound as bad as it did before, but the strange thing is that the noise happens just as often with 27 [osc~] objects as it did with 4374 [osc~] objects. That makes me wonder if there is something more going on here than just the number of [osc~] objects.

I've started looking at the poly thread, but it will take me a while to figure out whether there's something there that can solve my problem. Thanks.

@jamcultur For me 6400 copies of [osc~] seems to be the limit.
I have a vague recollection of hitting that limit before.
test-osc~.zip
David.

@whale-av Is there a way to route all of the [osc~] objects in all of the abstractions to a single [dac~] object?

@whale-av You can ignore my last question. I don't think a single [dac~] object would solve the problem. I'm looking at a patch I made last summer that has 2,940 [osc~] objects and 105 [dac~] objects and no noise problems. The logic is a bit different, but I don't yet see how the differences in the new patch could cause the noise. Time to do some more experiments...

@jamcultur You can [throw~ something] from [ton] and have a [catch~ something] in your main patch feeding into a [dac~]
It is strange that my test patch has more [osc~] objects than [tri_by_low] but I have removed [vline~] from [ton] and replaced it with [line] for the test patch.
David.

For examples of additive synthesis, take a look at D07.additive.pd or D13.additive.qlist.pd in Pd Vanilla's documentation.
Note-stealing yes! Turn silent instances dsp off with [switch~].
From a quick look: there is way too much running in parallel in your patch. Could be anything: objects and messages at the same time.
Also avoid clipping at sum of [dac~].

[osc~] or

[phasor~]
|
[cos~]

are reading from a cosine-wavetable.

Stacking many many lots of sines, it becomes cheaper to offline precompute a table for this and read that single wavetable with [tabosc4~], instead of many [osc~] or [cos~].

For Fourier-Synthesis, see sinesum message in helpfile of arrays:

[;arrayname sinesum 4096 1 0.5 0.25 0.125 0.0625, normalize 1(

Also you can do all sorts of offline computing signalerate~ objects, ether with
fast-forward message to Pd (see messages help-file)

or stepping audio-blocks
by banging [switch~] in a [pd sub-patch] (see [switch~] helpfile).

@lacuna Adding [switch~] to the abstraction almost completely eliminated the noise. Thanks! It didn't occur to me that [osc~] and [dsp~] objects that weren't producing sound would use so much system resources.

Next I'll look at wavetables. Can all the instances of an abstraction share the same wavetable, or does each instance need to have its own copy of the wavetable?