FLAC export shows lots of signs of clipping for a score that isn't too loud, WAV export works fine

[Apprisco]

Issue type

Import/export issue

Bug description

When exporting the same score on FLAC vs WAV, there is a clear difference in the tolerance for sound levels.
The sounds for this score aren't too loud at all- but FLAC is a unlistenable mess while WAV is absolutely perfect.
https://mega.nz/file/XtVmCQII#yI_-Z8ZaSCEd3CH8yTFVivV0_joMWyoANuGMU_M_cy0

Steps to reproduce

Make a somewhat loud score, preferably fff with -4dB or so.
Play some sounds.
Notice it's fine on playback. Notice the WAV export is fine.
Export with FLAC and notice crackles.
Up the dB value until issue is observed.

Screenshots/Screen recordings

https://mega.nz/file/XtVmCQII#yI_-Z8ZaSCEd3CH8yTFVivV0_joMWyoANuGMU_M_cy0

MuseScore Version

4.2.1, issue was ongoing since 2.0.

Regression

No.

Operating system

Windows 11

Additional context

No response

[Apprisco]

(Issue is at 0:48 of my piece)

[MikeRth]

The different behavior of FLAC export compared to WAV export should be easy to fix:

While WAV export handles clipping by cutting the wave at a maximum or minimum value (e.g. 32767 and -32768 when using signed 16 bits), FLAC export wraps because the overflow is not cut off at maximum and adding 1 to an integer value of 32767 results in -32768, so the waveform suddenly jumps to the other side of the range and continues there. The same goes for subtracting 1 from -32768 at the minimum.

This leads to a much more aggressive distortion than staying at maximum/minimum value. See attached waveforms of one of my MuseScore files exported to WAV (upper track) and to FLAC (lower track).

Overall, the best fix for this problem would be the proposal in "Indicate Clipping on Master Channel #17695" (#17695) probably combined with an option to automatically decrease the master fader level by the amount of clipping calculated. This should not be hard to implement if every audio data internally is calculated in floats between -1.0 and +1.0, as it is done in probably all DAWs nowadays.

It is really annoying having to load every exported file in Audacity to determine if clipping occurs. I generate audio demos for a choir and have to set the levels of the voices differently for each export (so that the loudest instrument can be clearly heard as the one for that choir voice demo). Because the instruments or voice sounds I use do not always behave identically, it could happen that it clips in one file but not in that for another voice, even when I use the same dB value.

A standard brickwall limiter for the output (real-time audio and exports) would be convenient for novice users too.