WIP A_Opus

[dericed]

partly resolves
#100

[dericed]

This needs a definition of the Initialization, ping @JeromeMartinez.

Here's an example of CodecPrivate in MediaTrace

000001D5    CodecPrivate (22 bytes)
000001D5     Header (3 bytes)
000001D5      Name:                                 9122 (0x023A2)
000001D7      Size:                                 19 (0x013)
000001D8     Identification (19 bytes)
000001D8      opus_codec_id:                        OpusHead
000001E0      opus_version_id:                      1 (0x01)
000001E1      channel_count:                        1 (0x01)
000001E2      preskip:                              312 (0x0138)
000001E4      rate:                                 48000 (0x0BB80)
000001E8      ouput_gain:                           0
000001EA      channel_map:                          0

[JeromeMartinez]

Look like Opus header is not required for decoding, only for metadata (i.e. channel mapping), defaults could be used (gain of 0? pre-skip from Matroska SeekPreRoll element) for other items.
Do we put Opus header mandatory or not?

[mbunkus]

Back when I added Opus support to mkvmerge I decided to use Xiph's Opus in Matroska mapping. The elements DiscardPadding (child of a BlockGroup), SeekPreRoll and CodecDelay (children of TrackEntry) were introduced for handling Opus' various requirements, and the Xiph wiki was later updated to name those new elements.

I strongly recommend to make the header mandatory. You could use defaults for the channels, but not for the output gain field of the Opus identification header.

[mbunkus]

To be clear: CodecPrivate stores the very same data structure called "Identification Header" as described in section 5.1 of RFC 7845.

[JeromeMartinez]

For the moment, as all other Matroska elements are already in the spec, I see only

Initialisation: The Private Data MUST contain the Identification Header packet, as defined in [RFC7845](https://tools.ietf.org/html/rfc7845), and no other data.

to add. Is it correct from your point of view?

[mbunkus]

It is, yes.

[dericed]

It setting Private Data to the Identification Header the only consideration. Should we also say

SampleFrequency MUST be set to 48000. SeekPreRoll MUST be set to 80000000. Channels MUST be set to the number of output PCM channels.

or address DiscardPadding, SeekPreRoll and CodecDelay in the Codec Mapping for Opus.

If the Private Data is the only consideration than the PR is updated for that.

[dericed]

Ping to @mbunkus @JeromeMartinez. I'm unsure what the status is here. Should we say that A_Opus infers some additional defaults on other Elements (IMHO codec-specific defaults makes for complex semantics).

[mbunkus]

Should we also say

SampleFrequency MUST be set to 48000. SeekPreRoll MUST be set to 80000000. Channels MUST be set to the number of output PCM channels.

or address DiscardPadding, SeekPreRoll and CodecDelay in the Codec Mapping for Opus.

No, we should not say that. Those values are valid for the current edition/version of the Opus specs. But the Matroska specs should not tie themselves to a certain version of a codec spec in such a way. What if Opus v3 decides to use a different pre-roll? We would have to adjust the Matroska specs.

If anything, we should say that SampleFrequency and SeekPreRoll be set according to the Opus version the track contains.

[mbunkus]

The description is meant to be about the mapping of the codec to Matroska: what blocks contain, how header fields have to be set (if anything special has to be observed) etc. It's not supposed to be a description of the codec itself. The Matroska specs are not codec specs or their marketing department.