Difference between revisions of "OggPCM"

From XiphWiki
Jump to: navigation, search
(added default channel mappings)
(s/subwoofer/LFE/; add left and right LFE for the crazy people.)
Line 75: Line 75:
This leaves two ways to define a  stereo file:
This leaves two ways to define a  stereo file:

Revision as of 01:54, 18 November 2005

This page was created as an alternative to the original OggPCM. The matter is currently a topic of heated debate and is now being voted on

Alternate format for OggPCM

The following is an alternate format for OggPCM. This is a work in progress and not a final proposal.

Note that unless otherwise noted, all multi-byte fields use the network byte order (big endian). The first packed in a stream MUST be the main header packet. The second packet MUST be the comment packet. Some extra header packets MAY be included after the comment header, provided this is identified in the main header. The packets that follow MUST all be data packets.

Main Header Packet

Multibyte fields in the header packets are packed in big endian order, to be consistent with network byte order. A header packet contains the following fields:

64  "PCM     " Codec identifier
16  0x00   Version Major (breaks backwards compatability to increment)
16  0x00   Version Minor (backwards compatable, ie, more supported format id's)
32  [uint] PCM format
32  [uint] Sampling rate [Hz]
8   [uint] Significant bits
8   [uint] Number of Channels (< 256)
16  [uint] Maximum number of frames per packet
32  [uint] Number of extra header packets

A PCM "frame" is composed of samples for all channels at a given time.

The "Codec identifier" is 64 bit long since most other Ogg codecs specify their identifier within the first 64 bits rather than the first 32 bits, so this allows applications to match on all 64 bits consistently.

The "Maximum number of frames per packet" field is meant to notify an application reading the file that no data packet will contain more than a certain number of frames. This not only makes implementation easier, but also provides information on how much needs to be buffered when streaming PCM files. A value of 0 means a maximum of 65536 frames. Implementations SHOULD make this field such that packets do not get split into multiple pages.

The "significant bits" field specifies how many bits are actually used. The other bits MUST be zero. This can be used to support audio with any resolution. For example, 12-bit PCM can be supported as "16 bit PCM" for the format and 12 for the number of significant bits.

For streams where the number of significant bits is the same as the bit width specified by the format, the significant bits field may be set to zero.

For streams where the number of significant bits is less than that specified by the bit width, the data shall be justified to fill the most significant bits. For 12 bit PCM in a 16 bit format, the 12 valid bits will occupy the 12 most significant bits of the 16 bit word and the least significant 4 bits shall be zero.

Since the main header packet and the comment packet are mandatory, the "extra header packets" field counts any additional header packets (aside from these two) that can be provided before the start of the data packets.

Comment packet

The codec header is followed by a "vorbis comment" packet and by optional extra headers, if any. The format used is the same as for Vorbis with the exception that there is no packet identifier (so the packet is exactly like it is for Speex).

Extra Headers (optional)

Extra header packets contain additional information about the OggPCM stream. Each extra header is defined as:

32  [uint] Header ID
...        Header data

Two examples for such optional extra header packets are the channel mapping and the channel conversion header packets.

Channel Mapping Header

The channel mapping header is defined as:

32 0x00000000   Header ID
16 [uint]   Major version
16 [uint]   Minor version
32 [uint]   Channel type
32xN [uint] Channel map (one for each channel)

All channel_types less than 0x80000000 are reserved for use by Xiph; 0x80000000 and above are allowed for application specific extensions.

This scheme allows for 2^31 -1 Xiph defined channel map types and 2^32 distinct channel names.

Exampe values for channel map type might be:


and defined channels might be:


This leaves two ways to define a stereo file:

channel_map [0] = OGG_CHANNEL_FRONT_LEFT
channel_map [1] = OGG_CHANNEL_FRONT_RIGHT

or as:

channel_map [0] = OGG_CHANNEL_FRONT_RIGHT
channel_map [1] = OGG_CHANNEL_FRONT_LEFT

Channel Conversion Header

Any number of channel conversion headers can be specified. This header specifies how to down-mix the data to another format.

32 0x00000001 Remixing Header Id
16 [uint]     Major version
16 [uint]     Minor version
32 [uint]     Target Channel type
32xMxN [sint] Target Channel (M) x Src Channel (N) Gain array

The ordering of the mixing matrix is such that source channel gains are consecutive. The gain (note: *signed* integer) has the 16 MSBs for the integer part (including sign) and 16 bits for the fracional part of the gain. Note: the gain can be negative.

Data Packets

Data packets contain the raw PCM audio in interleaved format (complete frames are encoded sequentially) with the following definitions/restrictions:

  • A PCM "frame" is composed of samples for all channels at a given time.
  • Any OggPCM packet MUST only contain complete frames (ie samples for all channels at a given sampling instance). Partial frames are forbidden.
  • There is no padding allowed in a frame except when some bits (<8) are needed to complete a byte. This means that packet size has a direct relationship to the number of frames in the packet (for purposes of seeking).
  • Recommended packet size is smaller than 4k since interleaving and seeking in Ogg bitstreams is done on the resolution of packets and thus larger packet sizes create suboptimal bitstreams.

Supported PCM Formats

 Format ID     Short Name             Description
 -- Integer coding
 0x00000000    OGGPCM_FMT_S8          Signed integer 8 bit
 0x00000001    OGGPCM_FMT_U8          Unsigned integer 8 bit
 0x00000002    OGGPCM_FMT_S16_LE      Signed integer 16 bit little endian
 0x00000003    OGGPCM_FMT_S16_BE      Signed integer 16 bit big endian
 0x00000004    OGGPCM_FMT_S24_LE      Signed integer 24 bit little endian
 0x00000005    OGGPCM_FMT_S24_BE      Signed integer 24 bit big endian
 0x00000006    OGGPCM_FMT_S32_LE      Signed integer 32 bit little endian
 0x00000007    OGGPCM_FMT_S32_BE      Signed integer 32 bit big endian
 -- Compressed PCM
 0x00000010    OGGPCM_FMT_ULAW        G.711 u-law encoding (8 bit)
 0x00000011    OGGPCM_FMT_ALAW        G.711 A-law encoding (8 bit)
 -- IEEE Floating point coding
 0x00000020    OGGPCM_FMT_FLT32_LE    IEEE Float [-1,1] 32 bit little endian
 0x00000021    OGGPCM_FMT_FLT32_BE    IEEE Float [-1,1] 32 bit big endian
 0x00000022    OGGPCM_FMT_FLT64_LE    IEEE Float [-1,1] 64 bit little endian
 0x00000023    OGGPCM_FMT_FLT64_BE    IEEE Float [-1,1] 64 bit big endian

Format IDs below 0x80000000 are reserved for use by Xiph and all the ones above are allowed for application-specific formats.

Format Defaults

(ideas by JMV, not yet approved by anyone else. Should be merged in respective header definition above if approved)

In order to simplify implementations when it comes to channel mappings, several defaults are defined when no extra header is present.

Channel Encoding

  • Files containing one channel are assumed to be plain mono files with:
channel_type = OGG_CHANNEL_MAP_MONO
channel_map [0] = OGG_CHANNEL_FRONT_CENTER
  • Files containing two channels are assumed to be stereo files with:
channel_map [0] = OGG_CHANNEL_FRONT_LEFT
channel_map [1] = OGG_CHANNEL_FRONT_RIGHT
  • Files containing four channels are assumed to be quadraphonic files with:
channel_map [0] = OGG_CHANNEL_FRONT_LEFT
channel_map [1] = OGG_CHANNEL_FRONT_RIGHT
channel_map [2] = OGG_CHANNEL_REAR_LEFT
channel_map [3] = OGG_CHANNEL_REAR_RIGHT

Channel Conversion

  • Stereo files SHOULD be converted to mono files by averaging the left channel and the right channel
  • Quadraphonic files SHOULD be converted to stereo files by averaging the corresponding rear and front channels.

Related Links

Short info about AC-3: http://www.mediatwins.com/en/support/kb_topic_11.html

AC-3 spec: http://www.atsc.org/standards/a_52a.pdf
Note: around p34/140 it appears to be how the channel mapping is encoded.

.wav extended headers for multi channel: http://www.microsoft.com/whdc/device/audio/multichaud.mspx

General surround info: http://www.surroundassociates.com/fqmain.html