m (OggPCM moved to OggPCM Draft1)
Revision as of 01:15, 22 February 2006
What is it
OggPCM is a pulse-code modulation (PCM) audio codec for Ogg. Similar to Microsoft's .wav or Apple's .aiff formats, it's a simple way to store and transfer uncompressed audio within an Ogg container. For the purposes of this document, the term PCM is used to describe a digital representation of an audio signal, where volume samples are taken at regular uniform intervals and then quantized into a digital (usually binary) code. A more complete definition of PCM and related terminology can be found at Wikipedia.
Why is it
The intention for this format is as an interchange format, for example for use with OggStream. It is also useful for storing time-synced decoded audio/video, as opposed to using RIFF/WAV (.wav) and YUV4MPEG (.yuv) in separate files as was done during Theora development. It is intended to be less complex to use than either RIFF or AIFF.
A stream is composed of a header packet, zero or more comment packets, and one or more data packets. Data packets may be of variable length, including zero. The only valid use of a zero length data packet is to mark the end of stream. Data packets must contain samples for all channels. That is to say, the length of a data packet must be a multiple of the number of channels times the storage size of a single sample. For instance, for a stream containing 6 channels at 2 byte per channel, the length of the data packet must be a multiple of 12 bytes.
The degenerate stream is a single header packet followed by the raw data packets. While this degenerate stream is not incredibly useful for long term storage or as a general purpose container, it is useful for applications where other data describing the stream is available out of band, for instance amongst cooperating applications in an inter-process communication scheme. Streams providing the extra defined comment packets are intended to be useful for long term storage and communication amongst diverse applications.
Header and comment packets are processed as per the value of their first byte. Packets of unknown ID should be silently ignored, providing a convient way to add future expandability which does not break the data format. An example of how this can be useful is the proposed ReplayGain extension to .wav format: http://replaygain.hydrogenaudio.org/file_format_wav.html
The header packet contains a field indicating the number of comment packets preceding the raw data. Applications must either parse or skip exactly this many packets, in addition to the header packet, before treating the stream as raw data.
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:
Packet 0, BOS, 16 bytes 8 0x00 Stream Header Packet ID 24 "PCM" Codec identifier - 8 0x01 Version Major (breaks backwards compatability to increment) 8 0x00 Version Minor (backwards compatable, ie, more supported format id's) 8 [uint] Number of header packets preceding data 8 [uint] Number of Channels, 0 = 256 - 16 [flag] Flags 16 [enum] PCM Format ID - 32 [uint] Sample Rate
The flags field is defined as follows:
Bit Description 15 (MSB) Interleaved/Chunked - If set, data in the packets is "chunked" by channel. In a data packet containing 3 channels and 2 samples/channel, the chunked storage order would be 001122. For the interleaved storage format (default), the order would be 012012. others ReservedApplications conforming to version 1.0 of this spec MUST:
- set all reserved flags to false (zero) when creating these streams.
- preserve all values of all reserved flags when reading or modifying these streams, unless the application sets the minor version field to zero, in which case the reserved flags must be set to false as well.
At this time, there is only one defined comment packet.
Comment Header Packet 8 0x01 Comment Header Packet ID 24 "PCM" Codec Identifier -- Continues as [Comment Header]
Data packets have no header word. This is done to preserve the alignment of the data payload. The contents of the data packets are specified by a combination of the 'PCM Format ID' field and the 'Flags' field. The length of the data packet must be a multiple of the number of channels specified in the header, and the storage size of a single sample, as specified by the 'PCM Format ID' field.
Supported PCM Formats
Formats are identified within a header packet by a 16 bit "format type" field. While most applications will treat this as an opaque type, it is possible to discern some information about the format from the value of this field itself. Specifically, the format's storage size, in bytes, and its byte ordering, can be discerned by parsing the lower 6 bits of the value. These values are exposed so that it is possible to extract individual samples without necessarily understanding the coding scheme involved. While for pratical purposes, due to performance concerns, most applications will choose to operate on a buffer directly, it is nonetheless possible to work a sample at a time.
Binary Value Meaning ..xxxx00 N/A, or data not accurately described by this scheme. ..xxxx01 Least significant byte first. Bytes are MS bit first. ..xxxx10 Most significant byte first. Bytes are MS bit first. ..xxxx11 Data is machine endian ..0000xx Data can not be described by this bytepacking scheme. ..0001xx Samples are stored using one byte per sample ..0010xx Samples are stored using two bytes per sample ..0011xx Samples are stored using three bytes per sample ..0100xx Samples are stored using four bytes per sample ..1000xx Samples are stored using eight bytes per sample
The remaining 10 bits describe the coding scheme used to convert the digital value to an audio signal. The following formats are defined for version 1.0 of this format. For purposes of attribution, it should be noted that these formats are the PCM formats supported by the Advanced Linux Sound Architecture (ALSA) project, and should be fairly comprehensive.
Format ID Short Name Description -- Signed integer coding (0) 0x0004 OGGPCM_FMT_S8 Signed integer 8 bit 0x0009 OGGPCM_FMT_S16_LE Signed integer 16 bit little endian 0x000A OGGPCM_FMT_S16_BE Signed integer 16 bit big endian 0x000B OGGPCM_FMT_S16 Signed integer 16 bit machine endian 0x000D OGGPCM_FMT_S24_3LE Signed integer 24 bit little endian 0x000E OGGPCM_FMT_S24_3BE Signed integer 24 bit big endian 0x0011 OGGPCM_FMT_S32_LE Signed integer 32 bit little endian 0x0012 OGGPCM_FMT_S32_BE Signed integer 32 bit big endian 0x0013 OGGPCM_FMT_S32 Signed integer 32 bit machine endian -- -- Unsigned integer coding (1) 0x0044 OGGPCM_FMT_U8 Unsigned integer 8 bit 0x0049 OGGPCM_FMT_U16_LE Unsigned integer 16 bit little endian 0x004A OGGPCM_FMT_U16_BE Unsigned integer 16 bit big endian 0x004B OGGPCM_FMT_U16 Unsigned integer 16 bit machine endian 0x004D OGGPCM_FMT_U24_3LE Unsigned integer 24 bit little endian 0x004E OGGPCM_FMT_U24_3BE Unsigned integer 24 bit big endian 0x0051 OGGPCM_FMT_U32_LE Unsigned integer 32 bit little endian 0x0052 OGGPCM_FMT_U32_BE Unsigned integer 32 bit big endian 0x0053 OGGPCM_FMT_U32 Unsigned integer 32 bit machine endian -- -- IEEE Floating point coding (2) 0x0091 OGGPCM_FMT_FLT_LE IEEE Float (-1,1) 32 bit little endian 0x0092 OGGPCM_FMT_FLT_BE IEEE Float (-1,1) 32 bit big endian 0x0093 OGGPCM_FMT_FLT IEEE Float (-1,1) 32 bit machine endian 0x00A1 OGGPCM_FMT_FLT64_LE IEEE Float (-1,1) 64 bit little endian 0x00A2 OGGPCM_FMT_FLT64_BE IEEE Float (-1,1) 64 bit big endian 0x00A3 OGGPCM_FMT_FLT64 IEEE Float (-1,1) 64 bit machine endian -- -- IEC958 coding (?) (3) 0x00CD OGGPCM_FMT_IEC958_3LE IEC958 Subframe, 24 bit little endian 0x00CE OGGPCM_FMT_IEC958_3BE IEC958 Subframe, 24 bit big endian 0x00D1 OGGPCM_FMT_IEC958_LE IEC958 Subframe, 32 bit little endian 0x00D2 OGGPCM_FMT_IEC958_BE IEC958 Subframe, 32 bit big endian 0x00D3 OGGPCM_FMT_IEC958 IEC965 Subframe, 32 bit machine endian -- -- Mu-Law coding (4) 0x0104 OGGPCM_FMT_MU_LAW Mu-Law -- -- A-Law coding (5) 0x0144 OGGPCM_FMT_A_LAW A-Law -- -- ADPCM coding (6) 0x0180 OGGPCM_FMT_ADPCM Ima-ADPCM -- -- GSM coding (7) 0x01C0 OGGPCM_FMT_GSM GSM -- -- 24 bit signed integer in 32 bit storage (8) 0x0211 OGGPCM_FMT_S24_LE Signed integer 24 bit little endian 0x0212 OGGPCM_FMT_S24_BE Signed integer 24 bit big endian 0x0213 OGGPCM_FMT_S24 Signed integer 24 bit machine endian -- -- 24 bit unsigned integer in 32 bit storage (9) 0x0251 OGGPCM_FMT_U24_LE Unsigned integer 24 bit little endian 0x0252 OGGPCM_FMT_U24_BE Unsigned integer 24 bit big endian 0x0253 OGGPCM_FMT_U24 Unsigned integer 24 bit machine endian -- -- 20 bit signed integer in 24 bit storage (10) 0x028D OGGPCM_FMT_S20_3LE Signed integer 20 bit little endian 0x028E OGGPCM_FMT_S20_3BE Signed integer 20 bit big endian -- -- 20 bit unsigned integer in 24 bit storage (11) 0x02CD OGGPCM_FMT_U20_3LE Unsigned integer 20 bit little endian 0x02CE OGGPCM_FMT_U20_3BE Unsigned integer 20 bit big endian -- -- 18 bit signed integer in 24 bit storage (12) 0x030D OGGPCM_FMT_S18_3LE Signed integer 18 bit little endian 0x030E OGGPCM_FMT_S18_3BE Signed integer 18 bit big endian -- -- 18 bit unsigned integer in 24 bit storage (13) 0x034D OGGPCM_FMT_U18_3LE Unsigned integer 18 bit little endian 0x034E OGGPCM_FMT_U18_3BE Unsigned integer 18 bit big endian -- Other coding schemes supported by ALSA but not specified here: MPEG -- TODO: ADPCM and GSM need further specification (or elimination) since these aren't really byte packed like the other formats here are.
Encapsulation in Ogg
Following standard terminology for uncompressed audio, an audio frame is the collection of samples for all channels for a single sampling period. For example, an audio frame for a stereo signal is a pair of sample values for the left and right channels.
The granulepos of an Ogg page indicates the presentation time of the last presentable element in the last complete packet within that page; for OggPCM, a granule is an audio frame. The granule position specified is the total audio frames in the stream including the last complete packet in a page. Audio frames must not be split across packets. The rationale here is that the position specified in the frame header of the last page tells how long the data coded by the bitstream is in samples as well as provides the current stream position to seeking routines. A truncated stream will still return the proper number of audio frames that can be decoded fully.