So Spotify normalizes both these files with a target of equivalent to -14 dB LUFS, according to the ITU 1770 standard.
- You have inaudible high-frequency content in your mix. Loudness algorithms (both ReplayGain and ITU 1770) do not have a lowpass cut-off filter, meaning any high-frequency content will add up to the energy measured by the algorithms and your track will be measured as louder by the algorithms than is actually perceived.
- You have a really loud master (true peaks well above -2 dB) which makes the encoding add some distortion, adding to the overall energy of the track. That’s the energy as perceived by the algorithm, which might be inaudible to you but adds to the loudness from the algorithm’s perspective.
- You’re not listening to a linear playback system. The ReplayGain algorithm (just like the ITU 1770 algorithm) can’t guess what audio playback system you’re using, so it can’t compensate for non-linearity in your system. Meaning, tracks that have more energy in the frequencies your system lifts up will sound much louder on your system.
- A track that is very dynamic but mastered to -14 dB LUFS will have its peak levels preserved when played on Spotify. If you compare that to a loudly mastered track, at – 6 dB LUFS for example, its peaks get lowered to – 8 dB LUFS. The two tracks will playback at the same perceived loudness level, but the loud or “peak” parts of the more dynamic track will be much louder.