dsi-muxer 1.0.0

Racjin PS2 DSI container multiplexer/demultiplexer

DSI Muxer

Multiplexer/demultiplexer for the Racjin PS2 DSI container format.

DSI is a proprietary streaming container used by Racjin for PS2 cutscenes in games like:

• Fullmetal Alchemist and the Broken Angel
• Fullmetal Alchemist 2: Curse of the Crimson Elixir
• Busin 0: Wizardry Alternative NEO

No DSI muxer existed before this tool. VGMToolbox could demux DSI files but not create them.

Install

pip install dsi-muxer

Usage

Command Line

# Extract video and audio from a DSI file
dsi-muxer demux M004.DSI --video M004.m2v --audio M004.adpcm

# Create a new DSI from video + audio
dsi-muxer mux --video subtitled.m2v --audio jp_audio.adpcm --blocks 307 -o M004_new.DSI

# Show DSI file info
dsi-muxer info M004.DSI

Python API

from dsi_muxer import DSI
from dsi_muxer.container import ensure_end_of_sequence

# Demux
dsi = DSI.from_file("M004.DSI")
video = dsi.extract_video()   # MPEG-2 elementary stream
audio = dsi.extract_audio()   # PS2 SPU ADPCM

# Mux with proportional audio (for perfect A/V sync)
video = ensure_end_of_sequence(video)
new_dsi = DSI.mux(video, audio, nblocks=307)
new_dsi.to_file("M004_new.DSI")

# Inspect
for block in new_dsi.block_info():
    print(f"Block {block['block']}: {block['frames']}f, "
          f"aud={block['audio_size']}, vid={block['video_size']}")

How It Works

The DSI Format

DSI files are divided into fixed 0x40000 byte (256 KB) blocks. Each block contains:

• A 32-byte header identifying two streams (audio + video) with type tags and sizes
• 32 bytes of zero padding
• Interleaved audio and video data

Audio uses PS2 SPU ADPCM (44100 Hz, stereo, 0x100 interleave). Video is MPEG-2 Main Profile (512x448, 29.97 fps).

Proportional Audio Muxing

The key to perfect A/V sync in DSI is proportional audio distribution:

audio_per_block = frames_in_block * (total_audio / total_frames)

Blocks with more video frames get more audio. This ensures each block's audio duration matches its video duration. The PS2's SPU2 buffer smooths per-block variation.

The video is byte-sliced as a continuous stream — GOPs flow freely across block boundaries, just like the original game files. No per-block re-encoding or extra sequence headers.

Why This Matters

Unlike Sony's standard PSS format (which has timestamps for A/V sync), DSI has no timestamps. Sync is entirely determined by the audio/video ratio per block. Getting this ratio wrong causes audible desync during PS2 playback.

Format Reference

Block Header (32 bytes, little-endian)

Offset	Type	Field
0x00	uint32	Stream count (always 2)
0x04	uint32	Stream 1 offset (always 64)
0x08	int32	Stream 1 type (-8192=audio, -16384=video)
0x0C	uint32	Stream 1 size
0x10	uint32	Stream 2 offset
0x14	int32	Stream 2 type
0x18	uint32	Stream 2 size
0x1C	uint32	Reserved (always 0)

License

MIT