Files

73 lines
4.6 KiB
Markdown

# Original default effects bank
Read-only EXE/OMT recovery and port SDL stream checks,2026-10-24. Text offsets
are VA minus0x400000. This is not original-game audio comparison or speaker testing.
Jimmy PostLoad423642 pushes string4EE08C=`SoundEffects.omt`, calls46A910,
and stores its database pointer to current sound database696008 at423656.
The separate sounds.omt has only4 waves and is not the numeric gameplay SFX bank.
SoundEffects.omt supplies251 Wave identities, including58=`door closing`,
59=`Door opening loop`,230=`winning prize2`.
458980 forwards three arguments to47D390;4589C0 forwards four to47D640 after
checking global4F8434. Both set volume through47DC80 from4F6B38.47D390 checks
current database696008 and tests/loads FourCC Wave/id via interface slots20/14
at47D3B6..DA. Contextual voice/music paths may temporarily switch the current
bank; the consumer currently uses the proven default for pickup/mechanism events.
This does not recover the entire original bank-switching/channel system.
Vertical-door417B28..43 calls4589C0(actor,-1,59,1), retaining its returned handle
while moving.41797E..4179AE stops that handle through458A00(handle,0), clears it
to-1 and calls4589C0(actor,-1,58,0). Reversing while the handle exists does not
start a second loop. Button411Cxx uses serialized AvailSound. Pickup435Fxx uses
serialized SoundIndex; trophy's230 identity is retained by the interaction path.
`tools/extractors/sound_bank.py` preserves each bounded source RIFF byte sequence,
archive/id/name/offset/hash and PCM duration in effects/effects.json. One resource,
162 bird1, contains format55 (MPEG Layer3) in RIFF. Its explicit fact sample count
is25344 at11025Hz; the encoded payload is retained unchanged. SDL3.4.2 local
src/audio/SDL_wave.c1755..57 explicitly rejects MPEG formats. Playback of162
therefore requires a separate decoder/conversion and remains unsupported; it is
not silently substituted. The currently connected lab effects use PCM resources.
PreparedEntityResources merges the bank with per-scene explicit references;
duplicate identities require equal duration and identical payload bytes. SceneAudio
uses independent one-shot streams and owner-keyed loops, stops door loops on
completion/reset, and disposes all voices on scene rebuild. Music and speech remain
independent. Original attenuation/pan/volume/channel limits and full jetpack sound
callbacks are incomplete, so current SDL mixing is an implementation policy.
Verification:2 Python tests compare all251 emitted RIFFs to exact original slices,
check original bank/call/FourCC instruction bytes, preserve source hash and reject
output in Original. Managed merge tests reject conflicting identities. The SDL
dummy-device --effects-audio-smoke consumes original58/59 streams, retains a
one-shot while stopping the loop, adds230 independently and clears all streams.
No physical speaker, original timing/position or native Windows audio claim follows.
## Cutscene music correction from user testing
The user reported music continuing during cutscenes in the packaged build.
The prior consumer always looped STRT music underneath all speech. Original CAM
activation416776..4167E1 explicitly checks current music database696004, obtains
Jimmy complete+0xA38 handle, calls47D7A0(handle,0), unloads his selected music
archive+0xA4C and clears696004. This happens before the new speech starts.
DEFAULT447D5C returns early if Jimmy already controls the player; otherwise
447F12..447F45 reloads his selected music and starts it again (source wave0).
SceneDirector now emits MusicStopped before an actual CAM speech and MusicRestarted
when DEFAULT returns control from a sequence. SceneAudio disposes the music stream
on stop and restarts the selected track on return; this is a restart, not a silent
advancing/resumed loop. Speech, effects and source scene actions remain separate.
The lab chain begins with CAM, so music stays stopped through its intervening MCA
shots until DEFAULT. A standalone MCA has its own source path431280; a universal
mute-on-any-speech rule is not established and has not been invented.
Seven exact EXE instruction checks guard the stop/restart sites. Director tests
check stop-before-speech, silence between chained shots, one natural restart,
no repeat restart with Jimmy already controlled, and skip restoration. The SDL
dummy --cutscene-audio-smoke runs the actual16-cue lab chain against original audio
resources and verifies zero queued music throughout, natural restart, reset silence
and skip restart. It passes. This fixes the implemented lab music lifecycle;
physical output/native Windows comparison remains user/runtime validation work.