ReplayGain is different for the stem version of the song

[cokeramirez]

Bug Description

I'm converting my songs to stems with a custom script using demucs (I did it with gemini) and I'm noticing the stem version has a wrong replaygain calculated see this example:

Left is the original song with a replaygain of -10dB, the stem version has -6dB calculated (wrong, too little reduction)

This is a picture of both files with their gain in 0:

You can clearly see the stem is way louder, so it's calculated replaingain is wrong

I manually matched the replay gain of the stem file:

and that's way closer (althought is not correct, stem is heard at a lower volume with the original file replaygain, but is way closer)

This is avideo of they playing with:

1.- the replaygain number igualated by hand,
2.- the gain matched by ear
3.- the stem with the recalculated replaygain value (way louder)

I recorded the window so it didn't show the option to clear the replaygain I selected

2026-05-27_17-45-15.mp4

Version

2.6-beta-281-g2a92f16cbf

OS

Windows 11

[daschuer]

No one is working in this. I move it to 2.6.1 to not block out 2.6.0 release.

[acolombier]

Hi, without the original file, it is hard to investigate. I have also converted (most of) my library to STEM and didn't notice anything of the sort. Could you please share your two files? Feel free to share them privately at antoine@mixxx.org