I redid Tequila’s FE8 walking sounds fix a while back (here) and documented the format as part of the installer.
The data is a 2-short header followed by the sound data. Sound data consists of song ID shorts or 0 to not play a sound that frame. The first short in the header is the number of sounds/lack of sounds in the sound data. The second short in the header is the distance between the high-priority version of a sound and its low-priority version in the song table. For example, the normal walking sounds have the following order in the song table:
...
96: walking sound 1 (medium priority)
97: walking sound 2 (medium priority)
98: walking sound 1 (mid-low priority)
99: walking sound 2 (mid-low priority)
...
and 98 - 96 = 2, hence the 2 in the header.