mirror/AC6_recomp
1
0
Fork 0
mirror of https://github.com/sal063/AC6_recomp synced 2026-06-15 06:31:24 -04:00
Code Issues Packages Projects Releases Wiki Activity

Added audio decompilation tool. Decodes only voicelines and BGM pack and movie packs for now. SFXs are stored under a different format not yet explored

Browse Source
This commit is contained in:
salh
2026-05-02 14:43:09 +03:00
parent 64f8efbb2b
commit 0fddc26760
2 changed files with 562 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/ac6_audio_extraction_walkthrough.txt
+299
View File
@@ -0,0 +1,299 @@
================================================================================
AC6 AUDIO EXTRACTION WALKTHROUGH
================================================================================
This walkthrough takes you from a fresh AC6_recomp checkout to playable .wav
files for every track of music, every voiceline, and every cutscene mix in
Ace Combat 6. It also explains why in-game sound effects (engines, missile
lock, explosions, weapons) are NOT covered yet, and what would be required to
extract them.
If you've never extracted assets from this repo before, do the texture
walkthrough first (docs/ac6_asset_extraction_walkthrough.txt). It gets the
runtime PAC dumps populated, which this walkthrough also benefits from.
================================================================================
0. WHAT YOU GET
================================================================================
After running this walkthrough you will have, in
out/ac6_runtime_audio/packs/ :
bgmpack__00000_off00000000.xma (47 BGM tracks; one .xma each)
bgmpack__00001_off01450000.xma
...
voicepack_eng__00000_off00000000.xma (7523 voicelines, English)
...
voicepack_jpn__00000_off00000000.xma (7523 voicelines, Japanese)
...
demopack_eng__00000_off00000000.xma (31 cutscene mixes)
...
If vgmstream is installed, each .xma also has a sibling .wav next to it.
You will ALSO have, in out/ac6_runtime_audio/, the runtime-captured FHM
RIFFs (currently 42 voice/event clips from the dumps you've already
collected).
You will NOT yet have:
- In-game SFX (engines, weapons, lock-on, explosions). See section 6.
- Cutscene video. moviepack.bin is standard ASF/WMV — see section 5.
================================================================================
1. PREREQUISITES
================================================================================
- The repo is built and ac6recomp.exe runs (texture walkthrough covers this).
- Python 3.10+ on PATH.
- Asset directory exists at:
out/build/win-amd64-relwithdebinfo/assets/
containing bgmpack.bin, voicepack_eng.bin, voicepack_jpn.bin,
demopack_eng.bin, demopack_jpn.bin, moviepack.bin.
- (Optional, for .wav decoding) vgmstream-cli.exe.
Download "vgmstream-win64-cli.zip" from https://vgmstream.org/downloads
and extract anywhere — e.g. C:\tools\vgmstream\.
================================================================================
2. THE QUICK PATH
================================================================================
From the repo root:
python tools/export_ac6_audio.py --packs ^
--vgmstream "C:/tools/vgmstream/vgmstream-cli.exe"
That's it. Output:
out/ac6_runtime_audio/ (FHM-channel RIFFs as .xma + .wav)
out/ac6_runtime_audio/packs/ (pack-channel streams as .xma + .wav)
Without vgmstream installed, omit --vgmstream and you'll get .xma files
only — no .wav. The .xma files are still the actual audio, just in the
Xbox 360 codec; any vgmstream-aware player (foobar2000 + vgmstream
plugin, vgmstream-cli later, etc.) will play them.
================================================================================
3. WHAT THE TOOL DOES, IN DETAIL
================================================================================
AC6 ships audio in two structurally different shapes. The exporter handles
each one differently.
----------------------------------------------------------------------------
3a) Standalone "pack" files in the asset directory
----------------------------------------------------------------------------
bgmpack.bin / voicepack_*.bin / demopack_*.bin are flat concatenations of
independent RIFF/WAVE streams, each aligned to a 0x800-byte sector.
Layout per pack:
+-------------------+ offset 0
| RIFF <size> WAVE | ----+
| fmt / x2st / data | | sub-stream 0 = 8 + declared_size bytes
+-------------------+ ----+
| (zero padding) | | aligned up to 0x800
+-------------------+
| RIFF <size> WAVE |
| ... | sub-stream 1
+-------------------+
...
The splitter reads the declared RIFF size at offset 4, copies
8 + declared_size bytes to its own .xma file, then advances to the next
sector boundary and looks for another RIFF. The codec inside is XMA2 for
bgmpack and XMA1 for voicepack/demopack — both decoded fine by vgmstream.
----------------------------------------------------------------------------
3b) FHM-embedded RIFF entries
----------------------------------------------------------------------------
When AC6 runtime-decompresses a PAC entry containing audio, the FHM
extractor (tools/extract_ac6_runtime_fhm.py) writes each audio child
as <NNN>_RIFF.bin. Those have the Project Aces RIFF/WAVE quirks:
- The RIFF size field is set to file_size, NOT file_size - 8. The tool
patches this on export.
- The fmt chunk uses wFormatTag = 0x0165 (XMA1) but actually carries
XMA2 streams (an "x2st" chunk follows). Decoders that key off
wFormatTag alone will misread channels/sample rate; vgmstream knows
to look at x2st.
The current corpus only has 42 RIFF entries, all from idx_0214 and
idx_0227 — these are voice/event clips. Adding more requires either
playing more of the game with AC6_DUMP_PAC_DECODED=1 or implementing the
offline mode-1 decompressor (see section 6).
================================================================================
4. COMMAND REFERENCE
================================================================================
Default (FHM corpus only, decode if vgmstream is on PATH):
python tools/export_ac6_audio.py
FHM corpus + asset packs:
python tools/export_ac6_audio.py --packs
Asset packs only, skip FHM corpus:
python tools/export_ac6_audio.py --packs-only
Dry-run (list what would be exported, write nothing):
python tools/export_ac6_audio.py --packs --dry-run
Skip the vgmstream decode step (write .xma only):
python tools/export_ac6_audio.py --packs --no-decode
Point at a specific vgmstream binary:
python tools/export_ac6_audio.py --packs ^
--vgmstream "C:/tools/vgmstream/vgmstream-cli.exe"
Custom asset directory (if your build output is elsewhere):
python tools/export_ac6_audio.py --packs ^
--assets "D:/games/AC6/assets"
================================================================================
5. CUTSCENE VIDEO (moviepack.bin)
================================================================================
moviepack.bin starts with the ASF GUID 30 26 B2 75 8E 66 CF 11 — it's a
standard Microsoft ASF container with WMV9 video and WMA audio inside.
Use ffmpeg directly (not vgmstream, not this tool):
ffmpeg -i moviepack.bin -map 0 -c copy out/movies/all.wmv
Or to remux into mp4 (re-encodes audio):
ffmpeg -i moviepack.bin -c:v copy -c:a aac out/movies/all.mp4
The ASF object index identifies individual cutscenes; ffmpeg's segment
muxer can split them, but moviepack also tends to be a single concatenated
ASF, in which case ffmpeg will emit one file. Use ffprobe to inspect.
================================================================================
6. THE MISSING SFX, AND WHAT IT TAKES TO GET THEM
================================================================================
If you extract everything above and play the .wavs, you'll hear: music,
dialogue, cutscene mixes. You will NOT hear: jet engines, missile lock
warnings, gun fire, explosions, ECM bursts, ground impacts, cockpit
warnings. Those live in a third channel that the current pipeline cannot
yet open.
Why:
- There is no sfxpack.bin. Look at the asset directory and you'll see
bgmpack / voicepack / demopack / moviepack — and that's it for audio.
- SFX waveforms are inside DATA00.PAC / DATA01.PAC, packed as RIFF/XMA
entries inside FHM containers like the voice clips above.
- Those FHM containers are mode-1 compressed. The runtime mode-1 hook
(see ac6_extraction_roadmap.md) decompresses entries as the game
requests them. Anything the game hasn't requested in your sessions
so far is still sitting in DATA*.PAC compressed.
- The cue tables that name the SFX are present already — idx_0000/
001_BSN_.bsn (sound bank), idx_0000/00{2,3,4}_BFX_.bfx (effects),
idx_0001/000_nusc.bin (Namco cue table) — but they only contain
pointers. The waveform data they point to is in the un-decompressed
SFX banks.
Two ways forward:
(a) Capture more at runtime.
Set AC6_DUMP_PAC_DECODED=1 and play missions with combat. The
SFX-bearing PAC entries will get decompressed by the game and
written to out/ac6_pac_runtime_dump/ as new entry_*_mode1_*.bin
files. Then re-run:
python tools/run_ac6_asset_pipeline.py --skip-pac-extract
python tools/export_ac6_audio.py --packs
New idx_<N> directories under out/ac6_runtime_fhm_typed/ should
contain new RIFF entries.
(b) Finish the offline mode-1 decompressor.
docs/ac6_extraction_roadmap.md "Path B" — port the guest mode-1
decoder to Python so every compressed PAC entry can be materialized
without launching the game. This is the only way to guarantee all
SFX are extracted, regardless of what mission state you've reached.
The recommended near-term option is (a). For a complete archive (b) is
the right answer.
================================================================================
7. TROUBLESHOOTING
================================================================================
"found 0 RIFF entries"
Your FHM corpus is empty. Run the asset pipeline first:
python tools/run_ac6_asset_pipeline.py
If that produces no output either, you don't yet have any runtime
dumps — see ac6_asset_extraction_walkthrough.txt.
"assets dir not found"
Pass --assets pointing at the directory that contains bgmpack.bin
etc. The default assumes a standard build at
out/build/win-amd64-relwithdebinfo/assets/.
"vgmstream-cli not on PATH; writing .xma only"
Not an error — the .xma files are valid. Install vgmstream
(https://vgmstream.org/downloads) or pass --vgmstream <path> and
re-run. vgmstream is also available as a foobar2000 plugin if you
just want to listen, not convert.
"truncated stream at 0x... ; stopping"
The pack walker found a RIFF whose declared size runs past the end
of the file. Usually means the pack file is partial / damaged.
Verify the file size against your install media.
vgmstream decode fails with "unknown format"
Confirm the .xma file starts with 'RIFF' and that bytes 8..12 are
'WAVE'. If they aren't, the splitter mis-stepped and the file is
junk. Try running with --no-decode and inspect the output bytes.
The .wav plays as silence / noise
For the 42 FHM-channel clips: if vgmstream warns about the format
tag, that's the 0x0165-but-actually-XMA2 quirk. vgmstream handles
it correctly when an x2st chunk is present. If you patched the
file by hand and removed x2st, restore from out/ac6_runtime_fhm_typed
and re-run the exporter.
================================================================================
8. WHERE THINGS LIVE
================================================================================
Inputs:
out/build/win-amd64-relwithdebinfo/assets/bgmpack.bin
out/build/win-amd64-relwithdebinfo/assets/voicepack_*.bin
out/build/win-amd64-relwithdebinfo/assets/demopack_*.bin
out/build/win-amd64-relwithdebinfo/assets/moviepack.bin
out/ac6_runtime_fhm_typed/idx_*/ (FHM-channel RIFFs)
Outputs:
out/ac6_runtime_audio/ (FHM-channel .xma/.wav)
out/ac6_runtime_audio/packs/ (pack-channel .xma/.wav)
Tools:
tools/export_ac6_audio.py (this walkthrough's tool)
tools/extract_ac6_runtime_fhm.py (FHM extractor)
tools/run_ac6_asset_pipeline.py (orchestrator)
Reference:
docs/ac6_extraction_roadmap.md ("Audio extraction"
section, channel 1/2/3)
docs/ac6_asset_pipeline.md (overall pipeline)
tools/export_ac6_audio.py
+263
View File
@@ -0,0 +1,263 @@
#!/usr/bin/env python3
"""Export AC6 audio to .xma (and optionally .wav via vgmstream-cli).
Two sources are handled:
1. Runtime FHM corpus (default).
The FHM extractor drops audio entries as `*_RIFF.bin`. Each is a Project
Aces RIFF/WAVE container with fmt (tag 0x0165 / XMA1), ALIG, x2st (XMA2
stream descriptor), data. The RIFF size field is non-standard (= file_size
instead of file_size - 8) and gets patched on export.
2. Monolithic asset packs (--packs).
bgmpack.bin / voicepack_*.bin / demopack_*.bin are flat concatenations of
independent RIFF/WAVE streams (XMA2 for BGM, XMA1 for voice/demo), aligned
to 0x800 sectors. We walk the declared RIFF size to split them.
Note: in-game SFX (engines, weapons, explosions) are NOT in these packs --
they live in BSN/BFX/nusc banks pointing into DATA*.PAC entries that aren't
runtime-decoded yet. Those need either the offline mode-1 decompressor or
gameplay capture with AC6_DUMP_PAC_DECODED=1.
"""
from __future__ import annotations
import argparse
import shutil
import struct
import subprocess
import sys
from pathlib import Path
REPO_ROOT = Path(__file__).resolve().parent.parent
DEFAULT_INPUT = REPO_ROOT / "out" / "ac6_runtime_fhm_typed"
DEFAULT_OUTPUT = REPO_ROOT / "out" / "ac6_runtime_audio"
DEFAULT_ASSETS = REPO_ROOT / "out" / "build" / "win-amd64-relwithdebinfo" / "assets"
PACK_FILES = [
"bgmpack.bin",
"voicepack_eng.bin",
"voicepack_jpn.bin",
"demopack_eng.bin",
"demopack_jpn.bin",
]
def split_pack(pack_path: Path, out_dir: Path) -> int:
"""Walk a concatenated multi-RIFF pack and write each stream as its own .xma.
Each stream begins with `RIFF`<u32 size>`WAVE`. The next stream starts at
`pos + 8 + declared_size`, padded up to a 0x800-byte sector boundary.
"""
SECTOR = 0x800
written = 0
blob = pack_path.read_bytes()
end = len(blob)
pos = 0
stem = pack_path.stem
while pos + 12 <= end:
if blob[pos:pos + 4] != b"RIFF" or blob[pos + 8:pos + 12] != b"WAVE":
# Skip to next sector and re-check; some packs have padding regions.
pos = (pos + SECTOR) & ~(SECTOR - 1)
continue
declared = struct.unpack_from("<I", blob, pos + 4)[0]
total = 8 + declared
if pos + total > end:
print(f" [{stem}] truncated stream at 0x{pos:x} (declared {declared}, "
f"only {end - pos - 8} available); stopping")
break
out_name = f"{stem}__{written:05d}_off{pos:08x}.xma"
(out_dir / out_name).write_bytes(blob[pos:pos + total])
written += 1
# Advance to next sector boundary.
next_pos = (pos + total + SECTOR - 1) & ~(SECTOR - 1)
if next_pos == pos:
break
pos = next_pos
print(f" [{stem}] {written} streams written")
return written
def find_riffs(root: Path) -> list[Path]:
return sorted(p for p in root.rglob("*RIFF.bin") if p.is_file())
def derive_output_name(riff: Path, root: Path) -> str:
rel = riff.relative_to(root)
parts = list(rel.parts)
# parts[-1] is e.g. "001_RIFF.bin"; drop suffix.
leaf = parts[-1].removesuffix(".bin")
# Tag with idx_NNNN and any FHM container directories on the path so names stay unique.
prefix_parts = [p.replace("_FHM_", "f") for p in parts[:-1]]
prefix = "__".join(prefix_parts)
return f"{prefix}__{leaf}.xma"
def patch_riff(blob: bytearray) -> tuple[bytearray, dict]:
"""Verify it's a RIFF/WAVE and fix the size field. Return (patched, info)."""
if blob[:4] != b"RIFF" or blob[8:12] != b"WAVE":
raise ValueError("not a RIFF/WAVE container")
file_len = len(blob)
declared = struct.unpack_from("<I", blob, 4)[0]
fixed_size = file_len - 8
if declared != fixed_size:
struct.pack_into("<I", blob, 4, fixed_size)
info = {
"declared_riff_size": declared,
"patched_riff_size": fixed_size,
"fmt_tag": None,
"has_x2st": False,
"data_size": None,
}
# Walk chunks for diagnostics.
off = 12
while off + 8 <= file_len:
tag = bytes(blob[off:off + 4])
sz = struct.unpack_from("<I", blob, off + 4)[0]
if tag == b"fmt ":
info["fmt_tag"] = struct.unpack_from("<H", blob, off + 8)[0]
elif tag == b"x2st":
info["has_x2st"] = True
elif tag == b"data":
info["data_size"] = sz
break
off += 8 + sz
if sz == 0:
break
return blob, info
def run_vgmstream(vgmstream: str, xma_path: Path, wav_path: Path) -> tuple[bool, str]:
try:
proc = subprocess.run(
[vgmstream, "-o", str(wav_path), str(xma_path)],
capture_output=True,
text=True,
timeout=120,
)
except FileNotFoundError:
return False, f"vgmstream not found: {vgmstream}"
except subprocess.TimeoutExpired:
return False, "vgmstream timed out"
if proc.returncode != 0:
return False, (proc.stderr or proc.stdout or "vgmstream failed").strip().splitlines()[-1]
return True, ""
def main() -> int:
ap = argparse.ArgumentParser(description=__doc__.splitlines()[0])
ap.add_argument("--input", type=Path, default=DEFAULT_INPUT,
help=f"FHM-typed root (default: {DEFAULT_INPUT})")
ap.add_argument("--output", type=Path, default=DEFAULT_OUTPUT,
help=f"Output directory (default: {DEFAULT_OUTPUT})")
ap.add_argument("--packs", action="store_true",
help="Also split bgmpack/voicepack/demopack from --assets.")
ap.add_argument("--packs-only", action="store_true",
help="Skip the FHM corpus and only split the asset packs.")
ap.add_argument("--assets", type=Path, default=DEFAULT_ASSETS,
help=f"Asset directory containing *.bin packs (default: {DEFAULT_ASSETS})")
ap.add_argument("--vgmstream", default=shutil.which("vgmstream-cli"),
help="Path to vgmstream-cli. If found on PATH it's used by default.")
ap.add_argument("--no-decode", action="store_true",
help="Just copy/patch .xma files; skip vgmstream decode.")
ap.add_argument("--dry-run", action="store_true",
help="List what would be exported and exit.")
args = ap.parse_args()
do_fhm = not args.packs_only
do_packs = args.packs or args.packs_only
if do_fhm and not args.input.is_dir():
print(f"input not found: {args.input}", file=sys.stderr)
return 2
riffs = find_riffs(args.input) if do_fhm else []
if do_fhm:
print(f"found {len(riffs)} RIFF entries under {args.input}")
pack_paths: list[Path] = []
if do_packs:
if not args.assets.is_dir():
print(f"assets dir not found: {args.assets}", file=sys.stderr)
return 2
pack_paths = [args.assets / n for n in PACK_FILES if (args.assets / n).is_file()]
print(f"found {len(pack_paths)} asset packs under {args.assets}")
if args.dry_run:
for r in riffs:
print(" ", derive_output_name(r, args.input), f"({r.stat().st_size} B)")
for p in pack_paths:
print(" pack:", p.name, f"({p.stat().st_size} B)")
return 0
args.output.mkdir(parents=True, exist_ok=True)
decode = not args.no_decode and args.vgmstream is not None
if not args.no_decode and args.vgmstream is None:
print("note: vgmstream-cli not on PATH; writing .xma only "
"(re-run with --vgmstream <path> to decode to .wav)")
decoded = 0
failed = 0
pack_written = 0
pack_xma_files: list[Path] = []
if do_packs:
pack_dir = args.output / "packs"
pack_dir.mkdir(parents=True, exist_ok=True)
for p in pack_paths:
print(f"splitting {p.name}...")
pack_written += split_pack(p, pack_dir)
pack_xma_files = sorted(pack_dir.glob("*.xma"))
if decode and pack_xma_files:
print(f"decoding {len(pack_xma_files)} pack streams via vgmstream...")
for xma_path in pack_xma_files:
wav_path = xma_path.with_suffix(".wav")
if wav_path.is_file() and wav_path.stat().st_size > 0:
decoded += 1
continue
ok, err = run_vgmstream(args.vgmstream, xma_path, wav_path)
if ok:
decoded += 1
else:
failed += 1
print(f" {xma_path.name}: {err}")
for riff in riffs:
out_name = derive_output_name(riff, args.input)
xma_path = args.output / out_name
blob = bytearray(riff.read_bytes())
try:
patched, info = patch_riff(blob)
except ValueError as exc:
print(f" skip {riff}: {exc}")
failed += 1
continue
xma_path.write_bytes(patched)
tag = info["fmt_tag"]
tag_str = f"0x{tag:04X}" if tag is not None else "?"
marker = "x2st" if info["has_x2st"] else "----"
print(f" {out_name} fmt={tag_str} {marker} data={info['data_size']}")
if decode:
wav_path = xma_path.with_suffix(".wav")
ok, err = run_vgmstream(args.vgmstream, xma_path, wav_path)
if ok:
decoded += 1
else:
failed += 1
print(f" decode failed: {err}")
print(f"done. fhm_riffs={len(riffs)} pack_streams={pack_written} "
f"wav_decoded={decoded} failures={failed}")
return 0 if failed == 0 else 1
if __name__ == "__main__":
raise SystemExit(main())