New comment by kcgen on void-packages repository

https://github.com/void-linux/void-packages/pull/31258#issuecomment-853391106

Comment:
> The enable-floats option could probably be added to Void's FluidSynth package (https://github.com/void-linux/void-packages/blob/master/srcpkgs/fluidsynth/template), but I do not know if it would be a suitable default outside of DosBox.

I highly recommend it across the board.  Floats are already 32-bits - which gives you 2^16 times finer resolution to quantify an amplitude level versus CD-quality 16-bit sound.

Even 24-bit audio is considered sufficient for professional-grade studio work. Leaving this as a double means samples are a whopping 64-bit; which is.. vastly overkill.   

The other benefit of 32-bit floats (versus full-doubles) is that SIMD'able operations can pack twice the data into a single operation, which can make  difference on CPU limited hardware like the Pi 2 and 3.  I make these two points in another thread and FluidSynth's author was also in agreement -  but can't remember where now.