I created 2 new versions, the difference is that, SIDB and SIDB3 add the value of each "SID" channel, and reads the value to be written from a 3x256 byte table and write AY volume values at the end of each phase, SIDB8 and SIDB9 write each "SID" channel volume value when it got.SIDB8 and SIDB9 has much less noise, the old solution would be better, because by the volume table the volume values are linear, but because the AY write is slow it sounds worse.The difference between SIDB8 and SIDB9 is that I made changes on samples on SIDB9, because AY volume is not linear, in SIDB8 tha samples remained linear, I think in most cases SIDB9 is better.
ps.Last Ninja file names are mixed.