OK - seems Im right about the "reserving for seemless switching) I designed 2 tones. One was a 4 partial which I held on the lowest note only- the other was 2 partial which i used to add polyphony. I got to 104 partials (the 4 low note plus 50 notes of the twin Osc patch) before I got note steeling, when I added the next I lost 2 partials of the low note, and the next I lost it all. That means I had 104 partials used before stealing. Using the 128 note (not 256 voice) figures, that means there are 24 notes (or 48 voices) being reserved for something else....which I can only assume is the seemless switching.
This test seems pretty conclusive (if not the reasons).
We have 128 voices of polyphony...which is 256 stereo partials (or 256 voices). This is the first thing Roland need to change (if its possible with firmware) - as using mono waveforms shouldnt use 2 voices.... it would only every need 1.
There are then 24 notes of polyohony being reserved by the system for something. While I/we dont know what - it would seem sensible to assume thats for the seamless switching. So making that switchable on/off would give the possibility to increase polyphony if needed.
That gives us currently 104 notes of useable polyphony. This is 104 PARTIALS though, NOT Notes. This is key when designing sounds, and using the sequencer. Using 4 partials in every tone would reduce you polyphony to 26 notes. The vast majority of sounds (certainly synth ones) would use at least 2 waveforms - and while some orchestral sounds may only use a single - the two are likely to balance, so its realistic to assume a working base of 52 notes (not worst case, but average use and a base line to work from).
Its also important to note the effect of decay. While FX (including reverb) are not counted to polyphony, having a note decay via the partials amp envelope means its still using notes of polyphony when it decays. The same will be true of PCM sounds such as pianos etc where the sample decays after release. For many of us this will greatly reduce the actual working polyphony.... again by perhaps half on average.
So - ultimately, while we have some control with the design of sounds, we dont have total control of how polyphony is reserved/used - and actually we only have somewhere around 26 notes worth as a baseline average that we can actually rely on.
Thats not a lot if your using layers, or indeed splits in the sequencer. If you layered 4 x 4 partial tones - or 4 2 partial tones where the PCMs or amp envs have a decay, you could only play 6 notes - after which you'd get note stealing....
At least if you could turn off seemless switching, and dynamically assign polyphony dependant on the waves being stereo or mono - you could actually up that to 16 notes (given the same 4 x 4 partial tones, or 4 x 2 partial tones with decaying amp envs or Samples). Even thats not GREAT, but at least thats reasonably workable.
