I tried it a couple of ways. My original idea, for the exact reasons you elude to, was to have the FIRST voice at a velocity of 100 and all subsequent voices at a velocity of 1. Then I came to realize that the first voice was not being interrupted...which at first gave me enthusiasm! But at a velocity of 1, a voice is silent (still active, but not audible). So....I changed my sequence to have all subsequent voices after the first one with a velocity of 20, which makes a voice lightly audible. At no time was the first voice being stolen.mojkarma wrote:@RonF
Have you assured that all notes you play in are at the same velocity?
A further thing comes into my mind and that is the stealing algorithm. The jp80 maybe doesn't cut of the oldest note but the quietest one. But if all notes are equally loud in that case maybe the algorithm cuts of the oldest one. That could be a reason why in some cases it doesn't cut of the first note.
Then my final trial, last night anyways, was to have ALL voices at a velocity of 30. To me, this was the most telling "test". At a velocity of 30, the voices are very soft, but audible....however, as the sequence plays (each voice on an increasing pitch across the keyboard, each starting a quarter beat apart, at 75 BPM, with some duplicate pitches in the mid spectrum to allow for such a high voice count), and voices are fired one after another, you can easily hear how the output of the JP80 evolves as the db increases with building intensity. At first, its very soft and smooth; by about 25 voices its built into a typical thick saw wave pad with nominal db; by 50 voices its loud; by 64 voices its about the peak of what you would play the JP80 in normal circumstances without crossing over into clipping the output stage, and at this point, the voices are all clearly audible with no artifacts. As voices increase from here, you start to detect escalating compression in the output stage which changes the character of the resulting "wave" being output. But you can still clearly hear all the voices (because voices are pitched across the frequency spectrum, its easy to detect that the first lowest notes are still playing). Somewhere around 80 voices, something else starts to happen....but again, its not dropping the lowest (oldest) voices played. At this point, the db of the resulting output has increased past a clipping stage, and the keyboard is doing *something* to manage it....because at this point the newest notes begin to dominate the lowest/oldest notes....not steal them, mind you....but dominate them in "presence". In other words, the db of the oldest notes are being "pushed" into the background. By the time I hit 84 notes in some cases.....96 in other cases (particularly with the reverb turned off)....it is no longer reasonably detectable what is happening to the oldest notes.
The reverb thing furthers my thought that this has something to do with output compression or limiting, or the way the "system" manages over-load. As you can imagine, reverb is only adding to the db at the output stage. It is quite possible if not likely that Roland employs some kind of output limiting or other output management....but I did not detect traditional voice stealing, certainly not sequential voice stealing, and what I did hear leaves room for debate as to exactly what it is.
So....at this point all I can do is report these findings. I am not drawing any conclusions....but I can tell you this....its not like the old days when we had 36 or 48 note polyphony and you could easily test the note stealing using this type of test. When you are over 64 voices, no matter what you do, it becomes exponentially more difficult to verify the status of discrete voices. After a certain point, what I hear is the oldest notes reducing in db, not cutting off suddenly as a new note is introduced.
BTW....this is what makes this test perhaps more reliable...because when you load up 12 or more voices on one key...you are employing a more coarse escalation of voice activation (12 at a time vs 1 at a time), which may cause a jump in output....which MAY sound "more like" voice stealing (or may in fact BE voice stealing, I am not claiming it isn't at this point), at the very least it makes it harder to detect what exactly is happening.
Again....I will test further this weekend.