The premise of the DAW using 64 bit fp internal data paths is that the original 24 bit resolution of the source is always preserved no matter how low one sets a fader anywhere on the mixing board.
Now, 24 bit fixed is a ton of dynamic range. It's not only bulletproof as a final consumer format, it makes recording raw tracks very forgiving. You could have a max peak at -36db (which would look like a flat line on the screen unless you zoomed the waveforms) and it would still have 18 bits resolution. (Still higher res than a 16 bit CD.)
Sometimes we want to preserve the signal all the way down into the decimal dust and not have to mind the fixed boundaries. Maybe not so much because we heard something funny with 24 bit fixed. More keeping the internal floating point signal unaltered so we don't have to fuss over critiquing it for quality preserved. Which may be within out perception bias. And more to prevent weird edge cases that can come up than anything else.
32 bit floating point is the first choice for that.
Technically a compromise vs the direct 64 bit fp internal signal. A much more reasonable file size though and so much absolute overkill it's hard to argue about anything. You could argue that 24 bit fixed holds such a complete dynamic range that there's no reason to even preserve the floating point. If some sonic element was buried over 100db down from the peaks... ain't no one ever hearing that without hearing damage from said peaks anyway!
But if you maybe had some low levels - or peaks that went over zero (which is allowed by floating point math) - the floating point format would preserve it fully. If there were over zero peaks, the data would be preserved and you could turn it down later and still have the natural peaks there.
So, 32 bit floating point is over the top data preservation free from fixed boundaries that clip. After that we pretty much conclude that 64 bit floating point would just be a waste of file space.
|