Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 17 additions & 2 deletions psprecomp/games/dissidia/docs/audio.md
Original file line number Diff line number Diff line change
Expand Up @@ -160,10 +160,25 @@ No per-action attribution was made.
option of the installed game (the installer wizard runs it when "Japanese voices (undub)" is switched on). It compares the undub disc's
`DATA/PACKAGE.BIN` with the installed one and extracts only audio that differs: the `voice/en/*` entries into `<work>/voice_jp/` (same
asset paths: the game asks for the English path and the loose-file layer serves the Japanese file), audio entries whose path is not in
`package_paths.txt` (RIFF / SSCF data) into `<work>/voice_jp/by_crc/<crc>.bin`, and the `DATA/MOVIE/*.PMF` that differ into
`<work>/movie_jp/`. Menus, text and other languages stay as they are. Checked on the EU image with the USA undub: 5,663 voice files
`package_paths.txt` (RIFF / SSCF data) into `<work>/voice_jp/by_crc/<crc>.bin`, and for the `DATA/MOVIE/*.PMF` that differ, the
installed movie with the undub's sound spliced in, into `<work>/movie_jp/`. Menus, text and other languages stay as they are. Checked on the EU image with the USA undub: 5,663 voice files
(177 MB), 316 by-CRC audio entries and 34 movies (488 MB); a second run extracts nothing new.

In the game: Settings > System > Voice (appears once `voice_jp/` exists; `voice=jp|en` in settings.ini, `PSPRECOMP_VOICE=jp|en`), applied at the
next start. The runtime side is `games/dissidia/runtime/hooks_io.cpp` (`extra_dirs` voice pack in `PackageMods::build`, `remap_disc_path`
for movies). Built and linked 2026-11-10; **not yet heard in game**.

**Movies keep the player's picture** (since 18 November 2026, `splice_movie_audio`). The USA undub has another picture in six of its 34
differing movies (`END`, `UP_END`, `UP_GIL`, `UP_OPN`, `UP_PRI`, `DES01_010`), so a whole copy showed it with Japanese voices on. A PMF is
2,048-byte packs of whole PES packets; the sound is private stream 1 (`0xBD`): a 4-byte header (stream number, where the first frame
starts) then 752-byte ATRAC3plus frames, contiguous from the first sound byte on both discs. So the undub's sound byte n goes where the
installed movie's sound byte n is, and every header stays right. `DES01_010`'s undub sound is one frame longer: the last frame is left
out. A movie whose streams do not have that layout is copied whole, as before. Method from VCKing101's port
(https://github.com/VCKing101/Dissidia012-pc, `docs/port/japanese-voices.md`, "Movies"). Checked on the EU disc with the USA undub:
all 34 have the installed picture byte for byte and the undub's sound byte for byte.

**Story scenes are not fixed.** The undub disc changes only sound files: no `talkevent/*.evex` scene script differs from retail. Each
line of a scene waits a fixed time tuned to its own release's recording, and 302 voiced scenes have other scripts on the Japanese disc
(55 with other voice lists, 247 with other waits; VCKing101, same doc, "Story scenes"). On the undub, as on a PSP running it, Japanese
lines play on the English scripts: some are cut off, a few play under the wrong line (012 Chapter 2, Vaan and Terra), and 013's
narrations mix English-only recordings into the Japanese ones. Fixing that needs the Japanese disc (NPJH-50377) for its scripts.
3 changes: 3 additions & 0 deletions psprecomp/games/dissidia/docs/related-projects.md
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,9 @@ top of 1 frame of delay covers about 100 to 130 ms without a stall. This is the

### 3. Japanese voices: story scene scripts

**Movies done** (18 November 2026, `tools/jp_audio.py`, [audio.md](audio.md)). The scenes are not: the undub disc changes no scene
script, so this needs the Japanese disc (NPJH-50377).

Source: [docs/port/japanese-voices.md](https://github.com/VCKing101/Dissidia012-pc/blob/main/docs/port/japanese-voices.md), "Story scenes".

Our `jp_audio.py` swaps audio files only. They found that **302 voiced story scenes** (`talkevent/*.evex`, ARC
Expand Down
78 changes: 73 additions & 5 deletions psprecomp/games/dissidia/tools/jp_audio.py
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,9 @@
<workdir>/voice_jp/<asset path> entries of the voice/en/* tree that differ (the game requests English paths; the
loose-file layer serves these instead when the Voice setting is Japanese)
<workdir>/voice_jp/by_crc/<crc>.bin audio entries (RIFF / SSCF data) whose path is not in package_paths.txt but that differ
<workdir>/movie_jp/<NAME>.PMF DATA/MOVIE movies of the undub disc that differ from the installed ones
<workdir>/movie_jp/<NAME>.PMF DATA/MOVIE movies whose sound differs: the installed movie's picture with the undub
disc's sound spliced in (splice_movie_audio; a whole copy of the undub movie only if
the two sound streams cannot be lined up)

Nothing else is copied (menus, text and the other languages stay the player's own). The runtime offers a "Voice" row in the
settings menu once <workdir>/voice_jp exists (games/dissidia/runtime/hooks_io.cpp). Re-running skips files already extracted.
Expand All @@ -29,6 +31,7 @@

SECTOR = 2048
AUDIO_MAGICS = (b"RIFF", b"SSCF")
ATRAC_FRAME = 752
PROGRESS = False


Expand Down Expand Up @@ -94,6 +97,53 @@ def file_md5(path):
return h.digest()


def _pes_audio(data):
"""[(start, size)] of the ATRAC3plus bytes of every private-stream-1 (0xBD) packet of a PMF, after each packet's 4-byte
header, and whether the frame starts that header gives fit 752-byte frames from the stream's first byte."""
off = struct.unpack(">I", data[8:12])[0]
out, pos, ok = [], 0, True
for p in range(off, len(data) - 13, SECTOR):
if data[p:p + 4] != b"\x00\x00\x01\xba":
continue
q, end = p + 14 + (data[p + 13] & 7), p + SECTOR
while q + 6 <= end and data[q:q + 3] == b"\x00\x00\x01":
length = struct.unpack(">H", data[q + 4:q + 6])[0]
if data[q + 3] == 0xBD:
s = q + 9 + data[q + 8]
first = struct.unpack(">H", data[s + 2:s + 4])[0]
n = q + 6 + length - s - 4
if n < 0 or (first != 0xFFFF and (pos + first) % ATRAC_FRAME):
ok = False
out.append((s + 4, n))
pos += max(n, 0)
q += 6 + length
return out, ok and pos % ATRAC_FRAME == 0


def splice_movie_audio(mine, undub):
"""The installed movie with the undub movie's sound: a PMF is 2,048-byte packs of whole PES packets, the sound private
stream 1 with a 4-byte header (stream number, where the first frame starts) and then 752-byte ATRAC3plus frames, from the
first sound byte on, on both discs. So the undub's sound byte n goes where the installed sound byte n is and every header
stays right. Where the undub's sound is longer (DES01_010.PMF, one frame) the rest is left out; where shorter, the
installed sound fills in. None when either stream does not have that layout. The picture is the player's own (the undub
disc's END, UP_END, UP_GIL, UP_OPN, UP_PRI and DES01_010 have another one). Same method as VCKing101's port
(https://github.com/VCKing101/Dissidia012-pc, docs/port/japanese-voices.md, "Movies")."""
a, ok_a = _pes_audio(mine)
b, ok_b = _pes_audio(undub)
if not (ok_a and ok_b and a and b):
return None
sound = b"".join(undub[s:s + n] for s, n in b)
out = bytearray(mine)
pos = 0
for s, n in a:
take = sound[pos:pos + n]
out[s:s + len(take)] = take
pos += n
if pos >= len(sound):
break
return bytes(out)


def main():
global PROGRESS
ap = argparse.ArgumentParser(description=__doc__.split("\n\n")[0])
Expand Down Expand Up @@ -193,13 +243,31 @@ def main():
hash_range(iso.f, r["extent"] * SECTOR, 0, r["size"]) == file_md5(mine_path):
continue
dest = os.path.join(out_movie, r["name"])
if os.path.isfile(dest) and os.path.getsize(dest) == r["size"]:
n_skip += 1
continue
spliced = None
if mine_path:
iso.f.seek(r["extent"] * SECTOR)
with open(mine_path, "rb") as mf:
spliced = splice_movie_audio(mf.read(), iso.f.read(r["size"]))
if os.path.isfile(dest):
# by content, not size: an earlier version copied the undub's movie whole, and UP_GIL's is the same size
if spliced is not None:
with open(dest, "rb") as df:
same = df.read() == spliced
else:
same = os.path.getsize(dest) == r["size"]
if same:
n_skip += 1
continue
n_movie += 1
if a.dry_run:
say(" would write movie %s (%d bytes)" % (r["name"], r["size"]))
say(" would write movie %s (%s)" % (r["name"], "your picture, Japanese sound" if spliced else "whole undub movie"))
elif spliced is not None:
os.makedirs(out_movie, exist_ok=True)
with open(dest + ".tmp", "wb") as o:
o.write(spliced)
os.replace(dest + ".tmp", dest)
else:
say(" %s: the sound streams do not line up, using the undub disc's whole movie" % r["name"])
copy_range(iso.f, r["extent"] * SECTOR, 0, r["size"], dest)
else:
say("the undub disc has no DATA/MOVIE folder: movies stay as they are")
Expand Down
78 changes: 78 additions & 0 deletions psprecomp/games/dissidia/tools/test_jp_audio.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
#!/usr/bin/env python3
"""Self-test for jp_audio.splice_movie_audio on synthetic PMF movies (no game data). Run: python3 -I test_jp_audio.py"""
import os
import struct
import sys

sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import jp_audio as ja

FRAME = ja.ATRAC_FRAME
HEADER = 0x800 # where the packs start (byte 8 of the file)


def pes(stream_id, payload):
hdr = b"\x81\x80\x00" # flags, no PTS, header length 0
return b"\x00\x00\x01" + bytes([stream_id]) + struct.pack(">H", len(hdr) + len(payload)) + hdr + payload


def movie(video, sound, chunk):
"""A PMF: per pack a video packet (from `video`) and a sound packet of up to `chunk` ATRAC bytes with its 4-byte header."""
out = bytearray(b"PSMF0015" + struct.pack(">I", HEADER))
out += b"\0" * (HEADER - len(out))
vpos = spos = 0
while spos < len(sound) or vpos < len(video):
pack = bytearray(b"\x00\x00\x01\xba" + b"\x44\x00\x04\x00\x04\x01\x01\x89\xc3\xf8")
v = video[vpos:vpos + 400]
vpos += len(v)
if v:
pack += pes(0xE0, v)
s = sound[spos:spos + chunk]
if s:
nxt = (-spos) % FRAME # bytes until the next frame starts
first = nxt if nxt < len(s) else 0xFFFF
pack += pes(0xBD, b"\x00\x00" + struct.pack(">H", first) + s)
spos += len(s)
pack += b"\xff" * (ja.SECTOR - len(pack))
out += pack
return bytes(out)


def sound_of(data):
parts, ok = ja._pes_audio(data)
return b"".join(data[s:s + n] for s, n in parts), ok


def main():
en_sound = bytes((i * 7) & 0xFF for i in range(FRAME * 5))
jp_sound = bytes((i * 13 + 5) & 0xFF for i in range(FRAME * 6)) # one frame longer, like DES01_010
mine = movie(b"E" * 3000, en_sound, 1500)
undub = movie(b"J" * 3200, jp_sound, 1600) # other picture, other packets
s, ok = sound_of(mine)
assert ok and s == en_sound, "parser: the installed movie's sound"
s, ok = sound_of(undub)
assert ok and s == jp_sound, "parser: the undub movie's sound"

out = ja.splice_movie_audio(mine, undub)
assert out is not None and len(out) == len(mine)
s, ok = sound_of(out)
assert ok and s == jp_sound[:len(en_sound)], "the undub's sound, its last frame left out"
parts, _ = ja._pes_audio(out)
masked_out, masked_mine = bytearray(out), bytearray(mine)
for st, n in parts:
masked_out[st:st + n] = masked_mine[st:st + n] = b"\0" * n
assert masked_out == masked_mine, "everything but the sound bytes is the installed movie (picture, packet headers)"

shorter = movie(b"J" * 3200, jp_sound[:FRAME * 4], 1600)
s, _ = sound_of(ja.splice_movie_audio(mine, shorter))
assert s == jp_sound[:FRAME * 4] + en_sound[FRAME * 4:], "a shorter undub sound: the installed sound fills in"

broken = bytearray(undub)
i = broken.index(b"\x00\x00\x01\xbd", HEADER + ja.SECTOR) # the second sound packet: a frame start off the 752 grid
broken[i + 9 + 2:i + 9 + 4] = struct.pack(">H", 5)
assert ja.splice_movie_audio(mine, bytes(broken)) is None, "a stream not on 752-byte frames is refused"
print("test_jp_audio: ok")


if __name__ == "__main__":
main()
Loading