Bit depth on recording vs. mp3 export

I use Audacity 2.1.3 in Windows 10, so far only for mono spoken word recording.

I think Gale Andrews somewhere wrote that Audacity internally converts all recording to 32-bit float. Correct? If not, please clarify. If yes:

Within a project:
– What effect on recording quality, file size, or whatever does the Edit > Preferences > Quality > Default sample format setting (16-bit, 24-bit, or 32-bit float) have when recording a track?
– If a track is recorded at one sample format setting but its format is changed once recorded, what effect does this have?

Exporting to mp3: The export menu allows choosing bit rate (e.g., 128 kbps) but not apparently bit format (16, 24, or 32). Is the same bit format used for all mp3 export regardless of that used for recording? If yes what is it? If not please clarify.

I just found the Audacity Wiki item on bit depth, which answers the first question about file size and recording quality. But I’d still appreciate replies to the questiosn about changing a track’s bit format in a project after recording, and about bit depth of exports to mp3. Thanks.

Audacity works internally in 32-bit float format.

If you have tracks that are 16 bit (BAD idea), then for each process that you apply to the track, Audacity will read the 16-bit data, convert to 32-bit, process it, convert back to 16-bit (with “dither” by default) and return 16-bit data back to the track.

By default, Audacity converts imported files to 32-bit when you import them, so that you then have 32-bit float data in the track, and no format conversion occurs until you export (MUCH better).

That sets the format that the track data will be when you record.
It also sets the format for ‘most’ imported file types, though some file types will import at 16 bit.

I assume that you mean changing the “Format” from the track drop down menu?
If so, then changing the format does what it says: it changes the format of the track data. For example, if you have a 16-bit track and change the track Format setting to 32-bit float, then each sample value will be converted from a 16-bit integer number to a 32-bit float number. Converting to a higher number of bits is essentially just adding padding, so “lossless”. Converting to a lower number of bits requires some form of “rounding”, so is imperfect.

MP3 doesn’t use “bit depth”. The concept of “bit depth” relates to “uncompressed” formats, typically “PCM” (see:
MP3 represents audio in a completely different way - it’s more like an “equation” which solves to an analog waveform.

“kbps” refers to the “bit rate” (kilobits per second = thousands of binary 1’s or 0’s per second).