format.md

relative	offset	absolute	size	description
----	0000	0000	000c	audio_container chunk header [C]
c	0000	000c	000c	audio_header chunk header [H]
ch	0000	0018	0004	size of track headers [T]
ch	0020	0038	0004	track count
ch	0050	0068	N*10	first stream name
chN	-	-	T*34	per-track data (see below)
chT	0000	-	N*10	second stream name
cH	0000	-	000c	audio_data chunk header
cH	000c	-	-	cluster sample data

Relative offsets are given as relative to the start (lower case) or end (upper case) of the named chunk or segment, with segment names in []'s following their description. For example, chT indicates an offset equal to the offset before the beginning of the audio_container contents (skipping 0xc bytes for the chunk header), plus the similar offset for the audio_header chunk, plus the named size "size of track headers". All numeric values above are given in hexadecimal.

A chunk header looks like:

    uint32 chunk_type
    uint32 chunk_size
    uint32 chunk_version // 0x1c020065 for DotD

Relevant chunk types are:

    0x0000080d audio_container
    0x0000080e audio_header
    0x0000080f audio_data

The information following the first stream name (relative offset ChN + 0x0) has the structure:

    [track_count] {
        uint32 unknown (identifier?)
        uint32 unknown (identifier?)
        uint32 unknown (identifier?)
        uint32 unknown (identifier?)
        uint32 unknown (zero)
        uint32 unknown (zero)
        uint32 track_byte_count
        uint32 track_byte_offset
    }
    [track_count] {
        uint32 track_total_used_bytes
    }
    [track_count] {
        uint32[4] unknown (checksum?)
    }
    [track_count] {
        char[8][] track_name
    }
    uint32 unknown (identifier?)
    uint32 unknown (identifier?)
    uint32 unknown (00 00 00 00)
    uint32 unknown (00 00 00 01)
    uint32 cluster_size
    uint32 unknown (identifier?)
    uint32 unknown (00 02 00 02)
    uint32 unknown
    uint32 cluster_used_bytes
    uint32 unknown
    uint32 sample_rate
    uint32 unknown
    uint32 unknown / first_track_sample_count?
    char unknown
    char num_channels
    uint16 unknown
    uint32[3] unknown
    uint32[4] unknown
    uint32 unused?
    uint32[4] unknown

(The notation [count] { ... } above indicates that the sub-structure in {}'s repeats N times, where N is the value of count. Typical byte values are in hexadecimal.)

The chunk headers are little endian. All other header data is big endian. Samples are also big endian. Track offsets are measured from the start of the audio_data chunk (following the chunk header for the same), i.e. an offset of 0 refers to the first byte of the first data cluster. Track counts include unused cluster bytes.

Samples are organized into clusters packed into the audio_data chunk with no further headers or additional padding beyond the cluster padding. Each cluster is cluster_size bytes long, of which only cluster_used_bytes contain sample data (the remainder are unused padding).

Tracks are composed of samples stored in clusters. Each track's clusters are contiguous, with the location and size described by track_byte_offset and track_byte_count, where the offset is relative to the start of the audio_data chunk data (i.e. with 0 being the first byte after the chunk header of the audio_data chunk). A track is assembled by reading up to cluster_used_bytes per cluster, up to track_total_used_bytes.