ALAC packets and elements

A packet is a bit stream read most significant bit first. It holds a sequence of elements, each beginning with a 3-bit type, and the last is an END element. A packet holds frameLength samples per channel, unless a partial-frame flag says otherwise.

Element types

Value	Name	Element	In the reference decoder
0	SCE	Single channel element	One channel
1	CPE	Channel pair element	Two channels
2	CCE	Coupling channel element	An error
3	LFE	Low-frequency effects element	One channel, read exactly as an SCE
4	DSE	Data stream element	Skipped
5	PCE	Program config element	An error
6	FIL	Fill element	Skipped
7	END	End of the packet	Aligns to a byte and stops

Elements for each channel count

Channels	Elements, in order
1	SCE
2	CPE
3	SCE CPE
4	SCE CPE SCE
5	SCE CPE CPE
6	SCE CPE CPE SCE
7	SCE CPE CPE SCE SCE
8	SCE CPE CPE CPE SCE

This is the order Apple’s encoder writes. The channels fill in the order of the layout tag, so a final SCE carries the low-frequency channel in the 5.1, 6.1 and 7.1 layouts.

SCE, LFE and CPE

These three share one layout. An SCE or LFE carries one channel; a CPE carries two, called U and V, and repeats the per-channel fields.

Bits	Field	Meaning
3	Element type	0, 1 or 3
4	Element instance tag	Identifies the element
12	Unused	Must be 0, or the decoder reports an error
1	partialFrame	1 means a sample count follows
2	bytesShifted	Whole low-order bytes split off each sample, 0 to 2. The value 3 is an error
1	escapeFlag	1 means the samples are stored uncompressed
32	numSamples	Only if partialFrame is 1: samples per channel in this packet, as two 16-bit halves, high first

Compressed data (escapeFlag 0)

Bits	Field	Meaning
8	mixBits	Fractional bits of the stereo matrix. Used by a CPE
8	mixRes	Signed stereo matrix weight; 0 means the channels are stored separately
4	modeU	0 runs the predictor once; anything else adds a first-order pass first
4	denShiftU	Fractional bits of the predictor coefficients
3	pbFactorU	Scales the entropy coder’s adaptation rate; 4 is the encoder’s default
5	numU	Number of predictor coefficients
16 × numU	coefsU	The signed coefficients
	modeV to coefsV	The same five fields for V, in a CPE only
varies	Shifted bits	Only if bytesShifted is not 0. See below
varies	Compressed samples	Entropy-coded predictor residuals for U, then for V

Apple’s decoder reads mixBits and mixRes for an SCE as well, and ignores them. Its encoder’s fast path writes mixBits 2 and mixRes 0, and uses 8 coefficients; its full search tries mixRes 0 to 4 and 4 or 8 coefficients. modeU is 0 in everything the encoder writes.

Shifted bits

Deep samples have their low bytes split off before prediction, because noise in the low bits does not compress. Apple’s encoder uses bytesShifted 1 for 24-bit audio, 2 for 32-bit and 0 otherwise. The bits come first, bytesShifted × 8 bits per sample, in sample order, with U and V interleaved in a CPE. After decoding, each sample is shifted left by that many bits and the stored bits are placed in the gap.

The number of bits predicted for each sample is bitDepth − bytesShifted × 8, plus 1 in a CPE to make room for the stereo matrix.

Uncompressed data (escapeFlag 1)

The samples follow directly at bitDepth bits each, with U and V interleaved in a CPE. There is no shifting and no matrix.

DSE

Bits	Field	Meaning
3	Element type	4
4	Element instance tag	Identifies the element
1	data_byte_align_flag	1 means the stream aligns to a byte before the data
8	count	Bytes of data. 255 means another 8-bit value is added to it
8 × count	Data	Skipped

FIL

Bits	Field	Meaning
3	Element type	6
4	count	Bytes of filler. 15 means an 8-bit value follows, and the count is 15 plus that value minus 1
8 × count	Filler	Skipped

END

Three bits, the value 7. The stream then aligns to the next byte, which ends the packet.

.M4A File Format – Packets and elements