I'm trying to understand why my output sometimes sounds lo-fi compared with professional productions, and that led me to ask if my use of 44.1/16 in most circumstances could be a factor. In other software I use, samples are processed as 32 bit floats, so I thought I might compare the output of paragraphic EQ when operating on a 16 bit sound file vs a 32 bit float sound file. I recorded a minute of voice at 44.1/16, saved it as Microsoft WAV, then opened it in a second data window. I then converted the bit depth of the 2nd data window to 32 (IEEE float) using Process->Bit Depth->Bit-Depth Converter. I applied paragraphic EQ to both files, 100% wet, 200 hz -Inf low shelf. To see if there was a difference, I copied the data in the 2nd window to a 3rd data window (which inherited the 32 bit float bit depth), then mixed the 1st window (still 16 bit) into the 3rd window, inverted. I expected the 3rd file to contain the sample-by-sample round-off errors, but Tools->Statistics shows that the entire file is composed of zeros. But if I re-save the original data window, save the 2nd data window, then reload both of them, now the errors show. So it seems like SF13 is also processing samples in some common high-precision bit depth, and only reflecting the configured bit depth when the data is saved to file.

So my question is: is it possible that I'm losing fidelity because I keep adding round-off error every time I save to disk? It's true that I tend to record, then save. Then normalize and save. Then de-noise and save, etc. All on different days. Can I assume that saving as a SF project won't help unless I use 32 bit float bit depth?