CSNG (File Format)

The CSNG format contains MIDI data. It appears in Metroid Prime 1 and 2. It is essentially MusyX's GameCube SNG music format, with a custom header.

Format

All offsets are relative to the start of the main header (after the custom header).

Overall, SNG functions similar to a Type-1 (multi-track) MIDI file, with (up to) 64 tracks merging their events into 16 General-MIDI sequencer channels. In addition to the multi-track structure, SNG supports storing MIDI regions as indexed patterns, so songs with repetitive refrains can save on memory by referencing patterns more than once.

Timings are represented in ticks. Unlike MIDI, the tick-resolution is fixed at 384 ticks per beat (e.g. 120 beats-per-minute works out to 384 * 120 / 60 = 768 ticks-per-second).

Custom Header

This 0x14-byte header isn't part of the MusyX format; it appears at the start of the file. After parsing this the rest of the file is copied into a buffer and then passed to the MusyX functions.

Offset	Size	Description
0x0	4	Magic; (always 0x2)
0x4	4	MIDI Setup ID
0x8	4	SongGroup ID
0xC	4	AGSC ID
0x10	4	SNG File Length
0x14	MusyX data starts

Header

Offset	Size	Description
0x0	4	Track Index Offset; usually 0x18 for GCN games
0x4	4	Region Data Index Offset
0x8	4	Channel Map Offset
0xC	4	Tempo Table Offset; 0x0 if tempo doesn't change
0x10	4	Initial Tempo; (commonly 0x78 or 120 beats per minute)
0x14	4	Unknown
0x18	End of header

Region Info

The track index has offsets relating the first region info for each track.

There is a sequence of at least 2 region info structures populating each track, last one acting as terminator:

Offset	Size	Description
0x0	4	Start Tick; time-point to begin executing region data
0x4	4	Unknown; commonly 0xffff0000
0x8	2	Region Data Index; used to select region data via the region data index; value is negative on a dummy region info to terminate track regions
0xA	2	Unknown
0xC	End of region info

Region Data

Here begins a free-form blob of indexed region data. It starts with a variable-length u32 array of SNG offsets for each region, then the region data itself.

Region Data Header

Offset	Size	Description
0x0	4	Region Data Header Size; size of the header after this field (always 0x8)
0x4	4	Pitch Wheel Data Offset; (absolute SNG-offset) 0x0 if no pitch-wheel messages in region
0x8	4	Mod Wheel Data Offset; (absolute SNG-offset) 0x0 if no mod-wheel messages in region
0xC	End of region data

Region Commands

After the region data header, the actual playback commands begin. There are only 2 types of commands in SNG: note and control change.

Delta Time RLE

Just like MIDI, each command starts with a delta time value telling the sequencer how many ticks to wait after the previous command. Unlike MIDI, Factor5 uses a custom RLE scheme to adaptively scale the value's precision to reduce the value's size.

The RLE operates on 16-bit words, with the value 0xffff triggering a continuation, then a 'dead' 16-bit word skipped over, then the 0xffff is summed with the following RLE value, looping the decode algorithm.

In Python, decoding works like so:

def DecodeDeltaTimeRLE(streamIn):
    total = 0
    while True:
        term = streamIn.ReadU16()
        if term == 0xffff:
            total += 0xffff
            dummy = streamIn.ReadU16()
            continue
        total += term
        return total

Note Command

When the two bytes following the delta-time != 0xffff, and the high-bit of the first byte is unset, this is a note command.

Unlike MIDI, which has separate commands for note-on/note-off, SNG attaches a note length value to a note-on command, which is then able to track its own lifetime.

Offset	Size	Description
0x0	1	Note; AND with 0x7f for the value
0x1	1	Velocity; AND with 0x7f for the value
0x2	2	Note Length; count of ticks before note-off issued by sequencer
0x4	End of note

Control Change Command

When the two bytes following the delta-time != 0xffff, and the high-bit of the first byte is set, this is a control change command.

See the standard MIDI control numbers for details.

Offset	Size	Description
0x0	1	Value; AND with 0x7f for the value
0x1	1	Control; AND with 0x7f for the value
0x2	End of control change

End Of Region

When the two bytes following the delta-time == 0xffff, this region has no more commands.

Continuous Pitch / Modulation Data

If the pitch or mod offsets in a region are non-zero, they point to a buffer of RLE-compressed (delta-tick, delta-value) pairs, decoding to signed 16-bit precision. The decoder must track the absolute time and value, summing each consecutive update for the current time/values.

The algorithm for this RLE is different than the delta-time one for commands. It may scale down to a single byte if able.

def DecodeRLE(streamIn):
    # high-bit shift-trigger RLE, limited to 2 bytes
    term = streamIn.ReadU8()
    total = term & 0x7f
    if term & 0x80:
        total = total * 256 + streamIn.ReadU8()
    return total

def DecodeContinuousRLE(streamIn, isValue):
    total = 0
    while True:
        # 1-2 byte RLE accumulated within continuable RLE
        term = DecodeRLE(streamIn)
        if term == 0x8000:
            total += 0xffff
            dummy = streamIn.ReadU8()
            continue
        total += term

        # values are signed deltas;
        # extending into the high-half of 15-bit precision
        if isValue:
            if total >= 0x4000:
                return total - 0xffff
            else:
                return total

        # times are always forward-deltas
        return total

Channel Map

This is a simple u8 table mapping 64 SNG tracks to 16 MIDI channels for instrument selection via the SongGroup MIDI-Setup.

Tempo Table

When the SNG has a non-zero tempo table offset, this song features tempo changes. The change events are simple absolute-tick / BPM pairs.

Offset	Size	Description
0x0	4	Tick; absolute time-point to perform tempo change
0x4	4	Tempo; new tempo in BPM
0x2	End of tempo change