Everything you hear here comes from running the cartridge's own sound engine (toca_el_sonido, 0x5C0E, and pide_un_sonido_sin_mirar_la_partida, 0x5BCA, which hands it the channels; the interrupt calls it at 0x4038) on an emulated Z80, frame by frame at 60 per second, just as the game calls it. What the engine writes to the PSG is turned into sound with an AY-3-8910 (tools/musica.py). Each MP3 is the whole piece up to where it starts again, plus 8 seconds of the repeat fading out; those that do not repeat end where the engine ends.
The PSG writes have been compared frame by frame with those of openMSX (C-BIOS MSX1 japonés). In every frame the PSG ends up the same:
The timbre is not checked: it is a textbook AY-3-8910, not a recording of the emulator. The probes are tools/omsx_psg.tcl and tools/coteja_psg.py.
pide_un_sonido takes the number in A and only sounds during a real game (bit 6 of 0xE002); the other entry skips that check. Below 0x87 it is a one-channel effect, from 0x87 to 0x89 it uses two and from 0x8A up all three. Tracks 0x86 and 0x87 are long entries in the table that no routine asks for with an ld a. The engine never writes the PSG mixer: start-up leaves it set, and here it starts from that same value, 0xB8.
The ending tune (0x8A)
Asked for by estado_12_fin_de_tacada (0x44F6); lasts 3.2 s.
The table is clear (0x88)
Asked for by la rutina de 0x4E60: la mesa se ha quedado sin bolas; lasts 2.4 s.
The exit jingle (0x8D)
Asked for by estado_8_confirma_la_seleccion (0x443C); lasts 2.8 s.
A long track in the table (0x86)
Asked for by ninguna llamada directa: está en la tabla de pistas; lasts 64.3 s.
Another long track in the table (0x87)
Asked for by ninguna llamada directa: está en la tabla de pistas; lasts 25.2 s.