Is this FM-X? I ask because FM-X polyphony is 1/2 AWM2.

Is this experiment using only one Part? Or more than one?

Operators used? Alg? Elements used?

... worst case in MODX with a single Part would be FM-X using all 8 operators. 8x1=8 for first note then +8 for next note = 16. ... It seems like even all 5 notes 8x5=40<64 (if a single FM-X Part was used - even maxed out).

I don't know what you have setup there to run the numbers. It's easiest to run the numbers with the algorithm with all 8 carriers:

[1] [2] [3] [4] [5] [6] [7] [8]

Because there's no ambiguity there between an oscillator and what would pull from polyphony.

I'm still not sure but currently assume carriers eat polyphony while modulators do not.

So

[1] [4]
[2] [5]
[3] [6] [7] [8] (which may not exist - but this shape) would assumed to only be able to "eat up" 4 units of polyphony per key simultaneous note triggered assuming the levels of the carriers are all non-zero.

AWM2 would have twice the headroom and a little easier to deal with since an element when sounding is like the all-carrier FM-X scenario above. Due to the relative difficulty in running into this issue with AWM2 - I'd assume FM-X. Not really sure, however, since the details are not here.

Jason, it's in the title.

it's AWM!!!

Which is why I'm surprised by this. It seems the programmers must have thought there was a shortcut to more efficient note stealing, in presuming that a certain number of repeated hits of the same note could earlier presume that they'd be able to drop/steal some of the old ones. I think it's a cache/pool, of about 5 or 6 tone plays on the same note before it kicks in.

Yeah - you'll have to excuse me at the moment. There's a good reason I missed the title even after thinking I rechecked to find it. No need to get into that.

Anyhow - for AWM2 how many Part(s) and elements?

Jason, sincerely, all the very best to you, and yours. Here's hoping and praying for you and them. Stay strong, good man!

In answer to your question: I found this when building a sound, from Init, with a single element on a single part.

Also, no effects, all dry. It's quite noticeable, but is an extreme scenario. I'm making a sound that's very dynamic to velocity, for lead play.

I'll build the same and see what I find here. Thanks for the well wishes.

I have gotten a lot of really good help and support from you, Jason, over the years, and I wish you all the best good fortune with whatever is going on.

I can confirm this behavior on Montage.

1) Recall "Init Normal (AWM-2)"
2) Edit element 1's AEG as follows:
2a) Release Time=42
2b) Time/Vel = -64
2c) Segment = Atk+Rls
3) Amplitude->Level/Pan: Level/Vel=+38 (I just like hearing the longer notes louder)

That's it. The basic rundown here is that the harder you strike the key - the more release you have.

This is the description of Time/Velocity from the parameter manual:

Negative values: High Velocities result in a slow AEG transition speed
while low Velocities result in a fast speed.

Indeed aggressive strikes of the key cause notes to ring for many, many seconds. Much longer than it takes to softly strike the next 4 notes. Soft notes release very fast and do not ring out.

Behavior described by Andrew (steps to reproduce here):

1) Configure Performance as above
2) [STORE] the Performance
3) Move data dial from [PEFORMANCE] (HOME) screen back/forth to advance/restore the Performance prepared for this test. This ensures no notes are ringing from previous testing (clean slate)
3) Strike note (say MIDI note D3) with high velocity which causes a ringing note. At this point a single element is sounding
4) Strike the same note (MIDI note D3) with low velocity 4 times staccato in rapid succession from the original note. Ensure rapid does not mean high velocity. This is more about the amount of time between notes - not the "speed" of the note press. It's easy to hear doing this correctly because the notes will not ring out and will have a short release.

After the 4th low-velocity note, the initial high-velocity note which is still releasing (not yet fully to a zero level) will immediately terminate.

This is not a polyphony cancellation - we're far below polyphony limits as described by values like "128". This is something else.

Modify the test and you will see this is related to how many times a single note can be "superimposed" over itself.

Assuming already having [STORE]'d the test vehicle for this ...
1) Move data dial from [PEFORMANCE] (HOME) screen back/forth to advance/restore the Performance prepared for this test. This ensures no notes are ringing from previous testing (clean slate)
2) Strike note (say MIDI note D3 as before) with high velocity which causes a ringing note. At this point a single element is sounding
4) Strike a different note (MIDI note D#3) with low velocity 4 (and more) times staccato in rapid succession from the original note. Ensure rapid does not mean high velocity. This is more about the amount of time between notes - not the "speed" of the note press. It's easy to hear doing this correctly because the notes will not ring out and will have a short release.

You will notice that striking a different note than the ringing note will not cause any interruption.

The conclusion here is that there is a 4-same-note limit for AWM2. I imagine something similar would happen if you use the sustain pedal and sustain something very loud then come back and play the same sustained note over 3 times softly - this will probably kick in the >4 same note limit and after the 3rd soft note - assuming these are all much softer - the louder sustained (1st) note would suddenly go away and the volume would overall drop unnaturally.

BTW: although there have been lots of polyphony topics - even ones that I think could be related to this discussion - I have not been able to quickly find any reference to exactly this observation. Thumbs up.

AWM2 polyphony seems to assume different notes. I'm not sure if this extends to different samples (where keybanks can have several keys sharing the same sample - pitch stretched). If each key of the default element in the initialized AWM2 has every key as a different sample then it would be interesting to find an instrument with less samples per key and see if that breaks worse (if all keys that share the same sample have the 4 max rule).

BTW: since notes have a maximum positive stretch - you can find the keybank breaks by shifting pitch with the modulation matrix by a maximum amount. It will max out the amount you can shift and each key that shares the same sample adds to this shift too. What you start to hear is the same shifted note until you get to the next sample.

In other words, you'll hear the same note when playing higher until you reach the next sample.

Experiments find this is not true:

AWM2 polyphony seems to assume different notes. I'm not sure if this extends to different samples (where keybanks can have several keys sharing the same sample - pitch stretched).

... at least different keys even when the keybank shares the same sample does not "group together". For the previous test, D3 and D#3 was used on the default "CF3 Stretch Sw St" waveform in the Init Normal (AWM2) Performance. CF3 Stretch SW St shares samples among D3, D#3, and E3. So even though I believe I was already testing same samples in the keybank (D3 + D#3) - when using different notes with the same sample this seemed to work out OK without the >4 rule.

In other words:

D3 (Sample XYZ)
+ 4x (or more) D#4 (Same Sample XYZ in keybank pitch stretched)
... did not stop the release of the D3.

The conclusion here is that the problem only manifests when the same MIDI note related is played simultaneously - and not connected at the sample level where multiple MIDI notes can share the same sample. Even two MIDI notes sharing the same sample will not manifest this behavior. And therefore - elements with very little samples (lots of pitch stretching) are just as safe as those with every key sampled. Seemingly no different in result. ... which is good considering.

This is the description of Time/Velocity from the parameter manual:
Negative values: High Velocities result in a slow AEG transition speed
while low Velocities result in a fast speed.

Sorry, I should have pointed this out, it's super unintuitive, and something I've internalised, so I forgot about it. SORRY you had to go down the rabbit hole to find this. It should be the other way around, like it is in all other multipliers.

The conclusion here is that there is a 4-same-note limit for AWM2. I imagine something similar would happen if you use the sustain pedal and sustain something very loud then come back and play the same sustained note over 3 times softly - this will probably kick in the >4 same note limit and after the 3rd soft note - assuming these are all much softer - the louder sustained (1st) note would suddenly go away and the volume would overall drop unnaturally.

And YES!!!

Now that I have come to know this as a thing, I can hear it when playing with pads and chords that I retrigger a lot in the way I sometimes build them up.

It's legion.

All very well written, Jason!

and THANK YOU, immensely, for confirming that it's not me. I use Unity every day, and find bugs, every day. Sometimes they're mine. But mostly they're not. Which gets a bit exhausting.

AWM2 polyphony seems to assume different notes. I'm not sure if this extends to different samples (where keybanks can have several keys sharing the same sample - pitch stretched). If each key of the default element in the initialized AWM2 has every key as a different sample then it would be interesting to find an instrument with less samples per key and see if that breaks worse (if all keys that share the same sample have the 4 max rule).

Is it possible to build a sound with a sample per note? eg I sample a lush analogue synth on every note... how do I (without using the drums thing) make a keyboard wide mapping of those sounds? Is this possible for end users to do?

Is it possible to build a sound with a sample per note?

Sample Robot at cost for MODX or the JM Tools Waveform Editor also at cost (for both MODX and Montage) would provide the means to do this.

That said - it doesn't affect this particular problem.

Those programs are the helpers. You can still load your own .WAV or .AIFF files (127 of them one-by-one) from a USB stick and create 127 single-note keybanks. Using only the touchscreen and your time. Less time if you only cover the region of the keyboard you can see (88-61 notes depending on your model) assuming no octave shifting or transposition.