registration is just all the things a panel memory sets up, mainly the combination of right and left hand sounds, but also a pattern, a split point, reverb and dsp setting etc etc.

If the song has used no edited sounds, load the song, then load the sound file using page 2 of the load page with everything switched off except the sound sub-file. Thus you are only overwriting the sound memory, and the song (sequencer), registrations (panel memories) and composer (possible pattern) are not affected.

Then press the panel memory you want to change, alter the sound to point to the new sound in the user memory, and resave the panel memory. If it is the first panel memory you may need to panel write the sequencer too.

If the song has already used edited sounds you don't want to overwrite these, so use page 3 of the load page to load only that individual (single) new sound to an unused sound memory position without overwriting the other used sound memories. Then resave the panel memory pointing to this sound memory position.

This is all assuming this easy recorded song has been made in a fairly standard fashion. Sight unseen there are so many variables it is impossible to give a definitive answer.