After not having touched my clone cards for some time, I recently got my hands on a more fitting CMOS oscillator, namely one with 2.4576 MHz instead of the 2 MHz I initially tried.
Feeding that clock frequency into the PSGs (instead of the approx. 2.3864 MHz of the halved ISA bus clock) yields audio output that is only off by almost exactly half a semitone.
That is less than the traditional “PAL shift” caused by broadcasting 24 FPS movies at 25 FPS and therefore essentially not noticeable, unless you are blessed (or in this case cursed) with perfect pitch hearing.

Well done. I'm not sure if i have perfect pitch, but as an amateur musician, i can easily detect even the slightest differences in pitch. I'm not obsessed by it though, in fact i even modified Tronix's AIL2 CMS driver effort to exclude the rhythm channel (seriously translating AIL2 rhythm channel notes into square waves sounds like ass) and lower the pitch by a handful of semitones much more to my liking.

I've got to look into programming these AY chips, maybe i can figure out an AIL2 driver for this standard.

To quote myself:

Benedikt wrote on 2025-09-15, 22:21:

There is also a matching new code path in https://github.com/planet-x3/px3_ose/tree/obscure-hw to effectively redirect Tandy sound to the MMB.

For a very quick and simple solution, that code could be transplanted into an audio driver.
Regarding the rhythm channel in that AIL2 CMS driver: Why did it not use one of the white noise channels? The CMS has two, doesn't it?