You are mixing up some things. Internally in a computer CPU, there are also 32bit streams, but that doesn’t mean that a 32-bit audio format has advantages. The best AD/DA convertors these days can only capture around 110 à 112dB of dynamic range. 24bit format gives you 144dB of dynamic range, and 32bit would give you 192dB. 192dB to capture what?
A 32-bit float engine converts incoming 32-int data to 32-float, and hence is lossy.
And it’s more precise than a 32bit fixed point calculator.
Really, you are mixing up internal CPU calculations/bit streams/data transport formats with audio file formats.
Bits = dynamic range. And since the highest dynamic range that can be captured is 112dB, there is absolutely no need for anything “bigger” than 24bit.
The ASIO driver converts the computer’s 1’s and 0’s to a fixed point audio format, and vice versa. So since the dynamic range of audio (input or output) can’t be higher than 112dB (limitation of ADC/DAC’s) why would you need anything “bigger” than 24 bit? Even if you would technically create a signal that has a dynamic range of 180dB, you would need to lower your master fader to not smoke the DAC, thereby bringing down the dynamic range (and bits) to what we have now.
Once the audio format is converted to 1’s and 0’s in the computer, which is nothing but a giant calculator, they are summed in 32-bit float (adding and substracting) and processed (EQ, Dynamics, etc …) in 64-bit float.
The errors that are occuring in a 32-bit float environment are smaller than the errors in a 32-bit fixed point environment, hence why fixed point calculations need dithering, and floating point calculations need randomizing.
Fredo