The magic cookie is ALAC’s configuration: everything the decoder must know before the
first packet. It is either 24 bytes (ALACSpecificConfig alone) or 48 bytes
(with channel layout information after it). Every field is big-endian, whatever the
container. A container stores the cookie unchanged and hands it to the decoder unchanged;
MP4 and CAF wrap it differently.
ALACSpecificConfig
| Offset | Size | Type | Field | Meaning |
|---|---|---|---|---|
| 0 | 4 | uint32 | frameLength | Samples per channel in a packet. 4096 is the default and the safe choice for compatibility |
| 4 | 1 | uint8 | compatibleVersion | Must be 0. The reference decoder refuses a higher value |
| 5 | 1 | uint8 | bitDepth | Bit depth of the source PCM, at most 32. The reference decoder handles 16, 20, 24 and 32 |
| 6 | 1 | uint8 | pb | Entropy coder tuning. Set to 40. Decoders use the value stored here |
| 7 | 1 | uint8 | mb | Entropy coder tuning. Set to 10 |
| 8 | 1 | uint8 | kb | Entropy coder tuning. Set to 14 |
| 9 | 1 | uint8 | numChannels | Channels, 1 to 8 |
| 10 | 2 | uint16 | maxRun | Set to 255. Currently unused |
| 12 | 4 | uint32 | maxFrameBytes | Largest packet in bytes; 0 means unknown |
| 16 | 4 | uint32 | avgBitRate | Average bit rate in bits per second; 0 means unknown |
| 20 | 4 | uint32 | sampleRate | Sample rate in hertz |
Apple’s document calls pb, mb and kb unused
tuning parameters that should be set to the values above, but the decoder does read them:
see the entropy coder.
Worked example
A 16-bit stereo stream at 44,100 Hz, with 4096 samples to a packet and the two sizes left unknown:
00 00 10 00 frameLength 4096
00 compatibleVersion 0
10 bitDepth 16
28 0A 0E pb 40, mb 10, kb 14
02 numChannels 2
00 FF maxRun 255
00 00 00 00 maxFrameBytes unknown
00 00 00 00 avgBitRate unknown
00 00 AC 44 sampleRate 44100Channel layout information
This optional 24-byte block follows the config. Without it, one channel means mono, two mean left and right, and more than two carry no particular roles.
| Offset | Size | Type | Field | Meaning |
|---|---|---|---|---|
| 0 | 4 | uint32 | channelLayoutInfoSize | 24 |
| 4 | 4 | fourcc | channelLayoutInfoID | chan |
| 8 | 4 | uint32 | versionFlags | 0 |
| 12 | 4 | uint32 | channelLayoutTag | One of the tags below |
| 16 | 4 | uint32 | reserved1 | 0 |
| 20 | 4 | uint32 | reserved2 | 0 |
A tag is a layout code in its high 16 bits and the channel count in its low 16. ALAC uses exactly one tag for each channel count:
| Channels | Tag | Name | Channel order |
|---|---|---|---|
| 1 | 0x00640001 | Mono | C |
| 2 | 0x00650002 | Stereo | L R |
| 3 | 0x00710003 | MPEG_3_0_B | C L R |
| 4 | 0x00740004 | MPEG_4_0_B | C L R Cs |
| 5 | 0x00780005 | MPEG_5_0_D | C L R Ls Rs |
| 6 | 0x007C0006 | MPEG_5_1_D | C L R Ls Rs LFE |
| 7 | 0x008E0007 | AAC_6_1 | C L R Ls Rs Cs LFE |
| 8 | 0x007F0008 | MPEG_7_1_B | C Lc Rc L R Ls Rs LFE |