For most practical purposes, the difference isn't very significant. Obviously, any recorded tracks have at most 24 bits of precision and only at the maximum levels. (Even then, the bottom 4 of the 24 bits is just random noise, intentionally.)
However, if the 32-bit file was the result of earlier processing (e.g., multi-track mixing but without normalizing), and if the file isn't normalized and the maximum level is very low, then you'd lose precision when doing any further processing or when sending to an audio interface (and bringing the volume up digitally.)
For example, if you have a mix of many tracks that was saved as 32-bit float, but with maximum level of say -24dBFS, and use SoX to discard all but a relatively quiet passage, and then amplify that passage (e.g., normalize it), you'd lose detail that was in the original recording. However, if you'd normalized that mix before saving it, then you'd lose very little when processing with SoX.
Another example where you'd lose some audio quality is when using a dynamic range compressor algorithm, for the same reason as selecting a quiet passage as mentioned above.
So, you'd not want to use SoX for normalization or substantial dynamic range compression.
This was interesting to me to learn. It's a minor limitation of SoX, but one to keep in mind. I use SoX as the last step or two, when preparing note samples for sampled musical instruments, so this isn't significant for my purposes.
Jeff
_______________________________________________ Sox-users mailing list Sox-users@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/sox-users
