Am 20.03.22 um 19:51 schrieb Ichthyostega: Beyond that, I also experimented with a different mixing curve for the crossfade (a classical S-shaped editing curve). I am not completely happy with the results, but at least it seems more like "turning over" to a new colour, and the marked decrease of volume during the transition is also reduced. (Another idea might be to use a different curve for the fade-out than for the fade-in?)> On Tue, 22 Mar 2022 03:37:02 +0100
Am 22.03.22 um 11:15 schrieb Will Godfrey:
This seems fine now. I can mess about as much as I like - and not an Xrun in sight. It also plays nice with Undo/Redo.
...and with the recent rework of the Legato/Portamento handling it now also works well with a heavy mix of Poly + Legato + Portamento, all while a long crossfade is underway.... :-D Moreover, I've sorted out the behaviour of the fade curve. As it turned out, an *equal power mix* is was does the trick here, since the changed wavetable will always re-randomise the phases. This implies that the old and the new waveform to be cross-mixed typically exhibit a low correlation, and thus we should keep the power, i.e. the square constant. Since taking the square root is somewhat expensive (and not easy to approximate much better than the SSE code does already), I now use my new exponential S-curve for the "stepping stones" (once per block) and do a linear interpolation in between. This way, we get a beautiful and unobtrusive transition, which does not "foreground" itself as a change. I will now re-base on top of current master and do some further testing. Next steps: - Investigate the timing regression (increased computation time on short PADsynth notes, which should not be the case). - GUI integration the "apply" button as discussed some time ago... -- Hermann _______________________________________________ Yoshimi-devel mailing list Yoshimi-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/yoshimi-devel
