SV8 specification draft

Nota : All fields, unless explicitely specified otherwise are read and written in Big-Endian order, independently of the underlying architecture.

File magic number

Magic number is on 32bits and is equal to 'MPCK' or 0x4D50434B
New value so old decoders will happily skip such files.

Block formatting

All blocks are formated using Key / Length / Value.
Key is 16 bits long. It's the block ID.
Length is a variable size field using the same values as ​nut :

bits, big-endian
0xxx xxxx                                           - value 0 to  2^7-2
1xxx xxxx  0xxx xxxx                                - value 0 to 2^14-2
1xxx xxxx  1xxx xxxx  0xxx xxxx                     - value 0 to 2^21-2
1xxx xxxx  1xxx xxxx  1xxx xxxx  0xxx xxxx          - value 0 to 2^28-2
...

The length is the size of the block in bytes, including the key and size fields. So the minimum length of a block is 3 bytes.
The value is the block content. Its size can be 0.
All unused bits in a block MUST be 0.

Field	length (bits)	Value
Key	16	"EX"
Length	n*8; 0 < n < 10	0x1A
Value	Length * 8	"example"

Summary of reserved block keys

Block name	Key	Mandatory
Stream information	SI	yes
Encoder information	EI	no
Audio data	AD	yes
Seek Pointer	SP	no
Seek Table	ST	no
Stream End	SE	yes

Stream information block

This block key is "SI".
This block contains the informations needed to decode the stream. This block is mandatory and must be written before the first audio block.

Field	length (bits)	Value	comment
CRC	32		CRC 32 of the block (this field excluded). 0 = invalid
Info length	n*8; 0 < n < 10		length (in bytes) from this field (included) to the ReplayGain version field
Stream version	8	0x08	bitstream version
Sample count	n*8; 0 < n < 10		number of samples in the stream. 0 = unknow
Sample frequency	4		see table below
Channel count	4	1..16	number of channels in the stream
Max used bands	5	1..32	maximum number of bands used in the file
MS used	1		MidSideStereo
Audio block frames	4	0..15	number of frames in an audio block is : 2Value (1..32768)
Unused	6		must be 0
ReplayGain version	8		The replay gain version - must be aligned on a byte -
Title gain	16		The loudness calculated for the title, and not the gain that the player must apply
Title peak	16
Album gain	16		The loudness calculated for the album
Album peak	16

Do we need to specify the channel position ? Is there standard positions / order of the channels ?

Frequency table

Value	Frequency (Hz)
0	44100
1	48000
2	37800
3	32000
4	96000

The CRC used is this one : ​http://www.w3.org/TR/PNG/#D-CRCAppendix.

The replay gain values are stored in dB in Q8.8 format (maybe Q7.9 or relative to the max value for the bitdepth is better ?).

Encoder infomations

This block key is "EI".

Field	length (bits)	Value	comment
profile used	4	0..15
PNS used	1
Unused	3		must be 0
MajorNumber	4
MinorNumber	8		even = release, odd = unstable
Implementation	4
Build	8

Audio data block

This block key is "AD".
This block contains audio frames. The first frame is a key frame.

Field	length (bits)	Value	comment
audio frames	?		n (or less if last block) frames of audio data as defined in SI block

Seek Pointer Block

This block key is "SP".
This block contains a pointer to the seek table block.

Field	length (bits)	Value	comment
pointer	n*8; 0 < n < 10		number of bytes from this block header to the "ST" block header

Seek Table Block

This block key is "ST".

Field	length (bits)	Value	comment
Seek Count	n*8; 0 < n < 10		number of seek elements in this table
Seek Distance	4	0..15	distance between referenced blocks = 2Value
Seek Datas	?

Format of seek datas :

• The reference for the seek size is the musepack magic number
  
• The first 2 values are stored using the same code as the block length code.
  
• The next values are coded as :
  

  code = value(n) - 2*value(n-1) + value(n-2)
  code <<= 1;
  if (code < 0)
    code = -code | 1;
  

code is sent as ​golomb code with M = 2¹².

Security Block

Checksum (MD5, SHA1) or error correcting code (LDPC).
To be defined later. May be better to keep security features external only.

End Block

This block key is "SE".
The recommanded block lenth is 3 bytes.
This block is mandatory and must be the last block of the stream. Tags are written after this block.

Tags

No block must be written after the end block to allow tagging by other applications.
Remarks about tagging? Incompatibilities?