In order to reduce/eliminate the accumulation of tiny processing errors, Audacity by default, does all of its intermediate processing in 32-bit floating point. This means that the final result before being exported will have a minimum of new errors introduced. There is no increasing of information by doing this - only a reduction of information lost.
Then tell him.
But of course, after doing most kinds of processing the original sample size only approximates the “true sample size”.
So rather than complain about the world why not be part of the solution?
That’s not the problem. If you increase the volume of a 16-bit file so it overloads, that is, the waveforms go over 0dB, that’s the end of the world. The performance is permanently damaged.
Audacity uses 32-Floating internally because it doesn’t overload. You can apply whatever filters, effects, or corrections you want and if the show gets too loud, just run an effect to reduce the volume. No clipping, no overloading, no harm done.
A real world application of this is Audiobook Mastering. It’s not unusual for the loudness tool to force expressive voice waveform peaks higher than 0dB. The next tool, the Limiter then gently brings the peaks back down to where they need to be with little or no sound damage.
The only shortcoming to all this is the need in most cases to go back to 16-bit for delivery to the client. That’s where you have the problem of more fidelity and higher accuracy inside Audacity than you can easily Export.
Unless, of course, the client can deal with 32-bit floating…
So is it not true that you cannot upsample? Meaning you can not increase the amount of information within an audio file once its recorded, you can only decrease it?
Right… You can’t add useful information, or useful resolution.
But of course you can up-sample the bit-depth and/or the sample rate. Re-sampling, both up-sampling and down-sampling is pretty common.
And as Koz says, you can temporarily make-use of the greatly expanded dynamic range during editing & processing. Floating point can go over +1000dB and below -1000dB. For all practical purposes it’s an infinite range… 16-bits has a range from 0 to -96dB (or maybe -93dB).
Behind the scenes, digital signal processing is just “easier” in floating-point. Virtually all audio processing software uses floating-point.
There are cases where you have to up-sample - For example, audio CDs are always 16-bits/44.1kHz and if you have a different format you’ll have to up-sample or down-sample. (Hopefully you don’t have to up-sample.) Or if you are mixing two different formats, one or both will have to be re-sampled. (Mixing is done by summation so you are increasing the “information”… If you sum two full-volume 16-bit files together it requires 17-bits.)
…I should mention [u]dither[/u]. It’s complicated but you’re “supposed” to add dither whenever you down-sample, and by default Audacity will dither when you export as 16 or 24-bits. That means if you open a 16-bit file, “do nothing”, and then export as 16-bits, dither (very slight noise) will be added so the audio data is altered.