draft-ietf-codec-oggopus.xml 66.6 KB
Newer Older
1
<?xml version="1.0" encoding="utf-8"?>
2
<!DOCTYPE rfc SYSTEM 'rfc2629.dtd' [
3 4 5 6 7 8 9
<!ENTITY rfc2119 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml'>
<!ENTITY rfc3533 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.3533.xml'>
<!ENTITY rfc3629 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.3629.xml'>
<!ENTITY rfc4732 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.4732.xml'>
<!ENTITY rfc5334 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.5334.xml'>
<!ENTITY rfc6381 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.6381.xml'>
<!ENTITY rfc6716 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.6716.xml'>
10
<!ENTITY rfc6982 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.6982.xml'>
11
]>
12 13
<?rfc toc="yes" symrefs="yes" ?>

14
<rfc ipr="trust200902" category="std" docName="draft-ietf-codec-oggopus-05">
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47

<front>
<title abbrev="Ogg Opus">Ogg Encapsulation for the Opus Audio Codec</title>
<author initials="T.B." surname="Terriberry" fullname="Timothy B. Terriberry">
<organization>Mozilla Corporation</organization>
<address>
<postal>
<street>650 Castro Street</street>
<city>Mountain View</city>
<region>CA</region>
<code>94041</code>
<country>USA</country>
</postal>
<phone>+1 650 903-0800</phone>
<email>tterribe@xiph.org</email>
</address>
</author>

<author initials="R." surname="Lee" fullname="Ron Lee">
<organization>Voicetronix</organization>
<address>
<postal>
<street>246 Pulteney Street, Level 1</street>
<city>Adelaide</city>
<region>SA</region>
<code>5000</code>
<country>Australia</country>
</postal>
<phone>+61 8 8232 9112</phone>
<email>ron@debian.org</email>
</address>
</author>

Ralph Giles's avatar
Ralph Giles committed
48 49 50 51 52 53 54 55 56 57
<author initials="R." surname="Giles" fullname="Ralph Giles">
<organization>Mozilla Corporation</organization>
<address>
<postal>
<street>163 West Hastings Street</street>
<city>Vancouver</city>
<region>BC</region>
<code>V6B 1H5</code>
<country>Canada</country>
</postal>
Ralph Giles's avatar
Ralph Giles committed
58
<phone>+1 778 785 1540</phone>
Ralph Giles's avatar
Ralph Giles committed
59 60 61 62
<email>giles@xiph.org</email>
</address>
</author>

Ralph Giles's avatar
Ralph Giles committed
63
<date day="15" month="October" year="2014"/>
64 65 66 67 68 69 70 71 72
<area>RAI</area>
<workgroup>codec</workgroup>

<abstract>
<t>
This document defines the Ogg encapsulation for the Opus interactive speech and
 audio codec.
This allows data encoded in the Opus format to be stored in an Ogg logical
 bitstream.
73 74 75 76
Ogg encapsulation provides Opus with a long-term storage format supporting
 all of the essential features, including metadata, fast and accurate seeking,
 corruption detection, recapture after errors, low overhead, and the ability to
 multiplex Opus with other codecs (including video) with minimal buffering.
77 78 79 80 81 82 83 84 85 86 87 88 89
It also provides a live streamable format, capable of delivery over a reliable
 stream-oriented transport, without requiring all the data, or even the total
 length of the data, up-front, in a form that is identical to the on-disk
 storage format.
</t>
</abstract>
</front>

<middle>
<section anchor="intro" title="Introduction">
<t>
The IETF Opus codec is a low-latency audio codec optimized for both voice and
 general-purpose audio.
90
See <xref target="RFC6716"/> for technical details.
91 92 93 94 95 96 97 98 99 100 101 102 103
This document defines the encapsulation of Opus in a continuous, logical Ogg
 bitstream&nbsp;<xref target="RFC3533"/>.
</t>
<t>
Ogg bitstreams are made up of a series of 'pages', each of which contains data
 from one or more 'packets'.
Pages are the fundamental unit of multiplexing in an Ogg stream.
Each page is associated with a particular logical stream and contains a capture
 pattern and checksum, flags to mark the beginning and end of the logical
 stream, and a 'granule position' that represents an absolute position in the
 stream, to aid seeking.
A single page can contain up to 65,025 octets of packet data from up to 255
 different packets.
Ralph Giles's avatar
Ralph Giles committed
104
Packets MAY be split arbitrarily across pages, and continued from one page to
105 106 107 108 109 110 111 112
 the next (allowing packets much larger than would fit on a single page).
Each page contains 'lacing values' that indicate how the data is partitioned
 into packets, allowing a demuxer to recover the packet boundaries without
 examining the encoded data.
A packet is said to 'complete' on a page when the page contains the final
 lacing value corresponding to that packet.
</t>
<t>
Ralph Giles's avatar
Ralph Giles committed
113
This encapsulation defines the contents of the packet data, including
114 115 116 117 118 119 120 121 122 123 124 125
 the necessary headers, the organization of those packets into a logical
 stream, and the interpretation of the codec-specific granule position field.
It does not attempt to describe or specify the existing Ogg container format.
Readers unfamiliar with the basic concepts mentioned above are encouraged to
 review the details in <xref target="RFC3533"/>.
</t>

</section>

<section anchor="terminology" title="Terminology">
<t>
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD",
126 127
 "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this
 document are to be interpreted as described in <xref target="RFC2119"/>.
128 129 130 131 132 133 134 135 136 137 138 139 140 141
</t>

<t>
Implementations that fail to satisfy one or more "MUST" requirements are
 considered non-compliant.
Implementations that satisfy all "MUST" requirements, but fail to satisfy one
 or more "SHOULD" requirements are said to be "conditionally compliant".
All other implementations are "unconditionally compliant".
</t>

</section>

<section anchor="packet_organization" title="Packet Organization">
<t>
142
An Ogg Opus stream is organized as follows.
143 144 145 146 147 148 149 150 151 152
</t>
<t>
There are two mandatory header packets.
The granule position of the pages on which these packets complete MUST be zero.
</t>
<t>
The first packet in the logical Ogg bitstream MUST contain the identification
 (ID) header, which uniquely identifies a stream as Opus audio.
The format of this header is defined in <xref target="id_header"/>.
It MUST be placed alone (without any other packet data) on the first page of
Ralph Giles's avatar
Ralph Giles committed
153
 the logical Ogg bitstream, and MUST complete on that page.
154 155 156 157 158 159 160 161 162 163 164 165
This page MUST have its 'beginning of stream' flag set.
</t>
<t>
The second packet in the logical Ogg bitstream MUST contain the comment header,
 which contains user-supplied metadata.
The format of this header is defined in <xref target="comment_header"/>.
It MAY span one or more pages, beginning on the second page of the logical
 stream.
However many pages it spans, the comment header packet MUST finish the page on
 which it completes.
</t>
<t>
166
All subsequent pages are audio data pages, and the Ogg packets they contain are
167 168
 audio data packets.
Each audio data packet contains one Opus packet for each of N different
Ralph Giles's avatar
Ralph Giles committed
169
 streams, where N is typically one for mono or stereo, but MAY be greater than
Ralph Giles's avatar
Ralph Giles committed
170
 one for multichannel audio.
171 172 173 174 175
The value N is specified in the ID header (see
 <xref target="channel_mapping"/>), and is fixed over the entire length of the
 logical Ogg bitstream.
</t>
<t>
176 177
The first N-1 Opus packets, if any, are packed one after another into the Ogg
 packet, using the self-delimiting framing from Appendix&nbsp;B of
178
 <xref target="RFC6716"/>.
Ralph Giles's avatar
Ralph Giles committed
179
The remaining Opus packet is packed at the end of the Ogg packet using the
180
 regular, undelimited framing from Section&nbsp;3 of <xref target="RFC6716"/>.
181 182 183 184 185 186 187
All of the Opus packets in a single Ogg packet MUST be constrained to have the
 same duration.
A decoder SHOULD treat any Opus packet whose duration is different from that of
 the first Opus packet in an Ogg packet as if it were an Opus packet with an
 illegal TOC sequence.
</t>
<t>
188 189 190
The coding mode (SILK, Hybrid, or CELT), audio bandwidth, channel count,
 duration (frame size), and number of frames per packet, are indicated in the
 TOC (table of contents) in the first byte of each Opus packet, as described
Ron's avatar
Ron committed
191
 in Section&nbsp;3.1 of&nbsp;<xref target="RFC6716"/>.
Ralph Giles's avatar
Ralph Giles committed
192
The combination of mode, audio bandwidth, and frame size is referred to as
193 194 195
 the configuration of an Opus packet.
</t>
<t>
196
The first audio data page SHOULD NOT have the 'continued packet' flag set
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
197 198
 (which would indicate the first audio data packet is continued from a previous
 page).
199 200 201 202 203
Packets MUST be placed into Ogg pages in order until the end of stream.
Audio packets MAY span page boundaries.
A decoder MUST treat a zero-octet audio data packet as if it were an Opus
 packet with an illegal TOC sequence.
The last page SHOULD have the 'end of stream' flag set, but implementations
Ralph Giles's avatar
Ralph Giles committed
204
 need to be prepared to deal with truncated streams that do not have a page
205 206
 marked 'end of stream'.
The final packet on the last page SHOULD NOT be a continued packet, i.e., the
Ralph Giles's avatar
Ralph Giles committed
207
 final lacing value SHOULD be less than 255.
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
There MUST NOT be any more pages in an Opus logical bitstream after a page
 marked 'end of stream'.
</t>
</section>

<section anchor="granpos" title="Granule Position">
<t>
The granule position of an audio data page encodes the total number of PCM
 samples in the stream up to and including the last fully-decodable sample from
 the last packet completed on that page.
A page that is entirely spanned by a single packet (that completes on a
 subsequent page) has no granule position, and the granule position field MUST
 be set to the special value '-1' in two's complement.
</t>

<t>
The granule position of an audio data page is in units of PCM audio samples at
 a fixed rate of 48&nbsp;kHz (per channel; a stereo stream's granule position
 does not increment at twice the speed of a mono stream).
It is possible to run an Opus decoder at other sampling rates, but the value
 in the granule position field always counts samples assuming a 48&nbsp;kHz
 decoding rate, and the rest of this specification makes the same assumption.
</t>

<t>
Ralph Giles's avatar
Ralph Giles committed
233
The duration of an Opus packet can be any multiple of 2.5&nbsp;ms, up to a
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
 maximum of 120&nbsp;ms.
This duration is encoded in the TOC sequence at the beginning of each packet.
The number of samples returned by a decoder corresponds to this duration
 exactly, even for the first few packets.
For example, a 20&nbsp;ms packet fed to a decoder running at 48&nbsp;kHz will
 always return 960&nbsp;samples.
A demuxer can parse the TOC sequence at the beginning of each Ogg packet to
 work backwards or forwards from a packet with a known granule position (i.e.,
 the last packet completed on some page) in order to assign granule positions
 to every packet, or even every individual sample.
The one exception is the last page in the stream, as described below.
</t>

<t>
All other pages with completed packets after the first MUST have a granule
 position equal to the number of samples contained in packets that complete on
 that page plus the granule position of the most recent page with completed
 packets.
This guarantees that a demuxer can assign individual packets the same granule
 position when working forwards as when working backwards.
For this to work, there cannot be any gaps.
</t>

Ralph Giles's avatar
Ralph Giles committed
257 258
<section anchor="gap-repair" title="Repairing Gaps in Real-time Streams">
<t>
Ralph Giles's avatar
Ralph Giles committed
259 260 261
In order to support capturing a real-time stream that has lost or not
 transmitted packets, a muxer SHOULD emit packets that explicitly request the
 use of Packet Loss Concealment (PLC) in place of the missing packets.
Ralph Giles's avatar
Ralph Giles committed
262
Only gaps that are a multiple of 2.5&nbsp;ms are repairable, as these are the
Ralph Giles's avatar
Ralph Giles committed
263 264
 only durations that can be created by packet loss or discontinuous
 transmission.
Ralph Giles's avatar
Ralph Giles committed
265 266
Muxers need not handle other gap sizes.
Creating the necessary packets involves synthesizing a TOC byte (defined in
Ralph Giles's avatar
Ralph Giles committed
267 268 269
Section&nbsp;3.1 of&nbsp;<xref target="RFC6716"/>)&mdash;and whatever
 additional internal framing is needed&mdash;to indicate the packet duration
 for each stream.
Ralph Giles's avatar
Ralph Giles committed
270 271 272 273
The actual length of each missing Opus frame inside the packet is zero bytes,
 as defined in Section&nbsp;3.2.1 of&nbsp;<xref target="RFC6716"/>.
</t>

Ralph Giles's avatar
Ralph Giles committed
274 275 276 277 278 279 280 281 282 283
<t>
Zero-byte frames MAY be packed into packets using any of codes&nbsp;0, 1,
 2, or&nbsp;3.
When successive frames have the same configuration, the higher code packings
 reduce overhead.
Likewise, if the TOC configuration matches, the muxer MAY further combine the
 empty frames with previous or subsequent non-zero-length frames (using
 code&nbsp;2 or VBR code&nbsp;3).
</t>

Ralph Giles's avatar
Ralph Giles committed
284 285 286 287
<t>
<xref target="RFC6716"/> does not impose any requirements on the PLC, but this
 section outlines choices that are expected to have a positive influence on
 most PLC implementations, including the reference implementation.
288
Synthesized TOC bytes SHOULD maintain the same mode, audio bandwidth,
Ralph Giles's avatar
Ralph Giles committed
289
 channel count, and frame size as the previous packet (if any).
Ralph Giles's avatar
Ralph Giles committed
290
This is the simplest and usually the most well-tested case for the PLC to
Ralph Giles's avatar
Ralph Giles committed
291 292
 handle and it covers all losses that do not include a configuration switch,
 as defined in Section&nbsp;4.5 of&nbsp;<xref target="RFC6716"/>.
Ralph Giles's avatar
Ralph Giles committed
293 294 295 296 297 298 299 300
</t>

<t>
When a previous packet is available, keeping the audio bandwidth and channel
 count the same allows the PLC to provide maximum continuity in the concealment
 data it generates.
However, if the size of the gap is not a multiple of the most recent frame
 size, then the frame size will have to change for at least some frames.
301 302
Such changes SHOULD be delayed as long as possible to simplify
 things for PLC implementations.
Ralph Giles's avatar
Ralph Giles committed
303 304 305 306 307 308
</t>

<t>
As an example, a 95&nbsp;ms gap could be encoded as nineteen 5&nbsp;ms frames
 in two bytes with a single CBR code&nbsp;3 packet.
If the previous frame size was 20&nbsp;ms, using four 20&nbsp;ms frames
Ralph Giles's avatar
Ralph Giles committed
309 310 311 312 313 314 315 316 317 318
 followed by three 5&nbsp;ms frames requires 4&nbsp;bytes (plus an extra byte
 of Ogg lacing overhead), but allows the PLC to use its well-tested steady
 state behavior for as long as possible.
The total bitrate of the latter approach, including Ogg overhead, is about
 0.4&nbsp;kbps, so the impact on file size is minimal.
</t>

<t>
Changing modes is discouraged, since this causes some decoder implementations
 to reset their PLC state.
319 320 321 322 323 324 325 326 327 328
However, SILK and Hybrid mode frames cannot fill gaps that are not a multiple
 of 10&nbsp;ms.
If switching to CELT mode is needed to match the gap size, a muxer SHOULD do
 so at the end of the gap to allow the PLC to function for as long as possible.
</t>

<t>
In the example above, if the previous frame was a 20&nbsp;ms SILK mode frame,
 the better solution is to synthesize a packet describing four 20&nbsp;ms SILK
 frames, followed by a packet with a single 10&nbsp;ms SILK
Ralph Giles's avatar
Ralph Giles committed
329 330 331
 frame, and finally a packet with a 5&nbsp;ms CELT frame, to fill the 95&nbsp;ms
 gap.
This also requires four bytes to describe the synthesized packet data (two
332
 bytes for a CBR code 3 and one byte each for two code 0 packets) but three
Ralph Giles's avatar
Ralph Giles committed
333
 bytes of Ogg lacing overhead are needed to mark the packet boundaries.
334
At 0.6 kbps, this is still a minimal bitrate impact over a naive, low quality
Ralph Giles's avatar
Ralph Giles committed
335 336 337 338
 solution.
</t>

<t>
339
Since medium-band audio is an option only in the SILK mode, wideband frames
340 341 342
 SHOULD be generated if switching from that configuration to CELT mode, to
 ensure that any PLC implementation which does try to migrate state between
 the modes will be able to preserve all of the available audio bandwidth.
Ralph Giles's avatar
Ralph Giles committed
343 344 345 346
</t>

</section>

347
<section anchor="preskip" title="Pre-skip">
348 349
<t>
There is some amount of latency introduced during the decoding process, to
350 351
 allow for overlap in the CELT mode, stereo mixing in the SILK mode, and
 resampling.
Ralph Giles's avatar
Ralph Giles committed
352
The encoder might have introduced additional latency through its own resampling
353
 and analysis (though the exact amount is not specified).
354 355 356 357 358 359 360 361 362 363 364
Therefore, the first few samples produced by the decoder do not correspond to
 real input audio, but are instead composed of padding inserted by the encoder
 to compensate for this latency.
These samples need to be stored and decoded, as Opus is an asymptotically
 convergent predictive codec, meaning the decoded contents of each frame depend
 on the recent history of decoder inputs.
However, a decoder will want to skip these samples after decoding them.
</t>

<t>
A 'pre-skip' field in the ID header (see <xref target="id_header"/>) signals
365
 the number of samples which SHOULD be skipped (decoded but discarded) at the
Ralph Giles's avatar
Ralph Giles committed
366
 beginning of the stream.
Ralph Giles's avatar
Ralph Giles committed
367
This amount need not be a multiple of 2.5&nbsp;ms, MAY be smaller than a single
368
 packet, or MAY span the contents of several packets.
Ron's avatar
Ron committed
369
These samples are not valid audio, and SHOULD NOT be played.
370
</t>
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390

<t>
For example, if the first Opus frame uses the CELT mode, it will always
 produce 120 samples of windowed overlap-add data.
However, the overlap data is initially all zeros (since there is no prior
 frame), meaning this cannot, in general, accurately represent the original
 audio.
The SILK mode requires additional delay to account for its analysis and
 resampling latency.
The encoder delays the original audio to avoid this problem.
</t>

<t>
The pre-skip field MAY also be used to perform sample-accurate cropping of
 already encoded streams.
In this case, a value of at least 3840&nbsp;samples (80&nbsp;ms) provides
 sufficient history to the decoder that it will have converged
 before the stream's output begins.
</t>

391
</section>
392

393
<section anchor="pcm_sample_position" title="PCM Sample Position">
394
<t>
Ralph Giles's avatar
Ralph Giles committed
395 396
<figure align="center">
<preamble>
397 398
The PCM sample position is determined from the granule position using the
 formula
Ralph Giles's avatar
Ralph Giles committed
399
</preamble>
400 401 402 403 404 405 406 407 408 409 410
<artwork align="center"><![CDATA[
'PCM sample position' = 'granule position' - 'pre-skip' .
]]></artwork>
</figure>
</t>

<t>
For example, if the granule position of the first audio data page is 59,971,
 and the pre-skip is 11,971, then the PCM sample position of the last decoded
 sample from that page is 48,000.
<figure align="center">
Ralph Giles's avatar
Ralph Giles committed
411 412 413
<preamble>
This can be converted into a playback time using the formula
</preamble>
414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433
<artwork align="center"><![CDATA[
                  'PCM sample position'
'playback time' = --------------------- .
                         48000.0
]]></artwork>
</figure>
</t>

<t>
The initial PCM sample position before any samples are played is normally '0'.
In this case, the PCM sample position of the first audio sample to be played
 starts at '1', because it marks the time on the clock
 <spanx style="emph">after</spanx> that sample has been played, and a stream
 that is exactly one second long has a final PCM sample position of '48000',
 as in the example here.
</t>

<t>
Vorbis streams use a granule position smaller than the number of audio samples
 contained in the first audio data page to indicate that some of those samples
Ralph Giles's avatar
Ralph Giles committed
434
 are trimmed from the output (see <xref target="vorbis-trim"/>).
435 436 437 438
However, to do so, Vorbis requires that the first audio data page contains
 exactly two packets, in order to allow the decoder to perform PCM position
 adjustments before needing to return any PCM data.
Opus uses the pre-skip mechanism for this purpose instead, since the encoder
Ralph Giles's avatar
Ralph Giles committed
439
 MAY introduce more than a single packet's worth of latency, and since very
Ralph Giles's avatar
Ralph Giles committed
440 441
 large packets in streams with a very large number of channels might not fit
 on a single page.
442
</t>
443
</section>
444

445
<section anchor="end_trimming" title="End Trimming">
446 447 448 449 450 451 452 453 454 455 456 457 458 459
<t>
The page with the 'end of stream' flag set MAY have a granule position that
 indicates the page contains less audio data than would normally be returned by
 decoding up through the final packet.
This is used to end the stream somewhere other than an even frame boundary.
The granule position of the most recent audio data page with completed packets
 is used to make this determination, or '0' is used if there were no previous
 audio data pages with a completed packet.
The difference between these granule positions indicates how many samples to
 keep after decoding the packets that completed on the final page.
The remaining samples are discarded.
The number of discarded samples SHOULD be no larger than the number decoded
 from the last packet.
</t>
460
</section>
461

462 463
<section anchor="start_granpos_restrictions"
 title="Restrictions on the Initial Granule Position">
464 465 466 467 468 469 470 471 472
<t>
The granule position of the first audio data page with a completed packet MAY
 be larger than the number of samples contained in packets that complete on
 that page, however it MUST NOT be smaller, unless that page has the 'end of
 stream' flag set.
Allowing a granule position larger than the number of samples allows the
 beginning of a stream to be cropped or a live stream to be joined without
 rewriting the granule position of all the remaining pages.
This means that the PCM sample position just before the first sample to be
Ralph Giles's avatar
Ralph Giles committed
473
 played MAY be larger than '0'.
474 475 476
Synchronization when multiplexing with other logical streams still uses the PCM
 sample position relative to '0' to compute sample times.
This does not affect the behavior of pre-skip: exactly 'pre-skip' samples
Ralph Giles's avatar
Ralph Giles committed
477
 SHOULD be skipped from the beginning of the decoded output, even if the
478 479 480 481 482 483 484
 initial PCM sample position is greater than zero.
</t>

<t>
On the other hand, a granule position that is smaller than the number of
 decoded samples prevents a demuxer from working backwards to assign each
 packet or each individual sample a valid granule position, since granule
Ralph Giles's avatar
Ralph Giles committed
485
 positions are non-negative.
486 487 488 489 490 491
A decoder MUST reject as invalid any stream where the granule position is
 smaller than the number of samples contained in packets that complete on the
 first audio data page with a completed packet, unless that page has the 'end
 of stream' flag set.
It MAY defer this action until it decodes the last packet completed on that
 page.
492 493 494 495 496
</t>

<t>
If that page has the 'end of stream' flag set, a demuxer MUST reject as invalid
 any stream where its granule position is smaller than the 'pre-skip' amount.
Ralph Giles's avatar
Ralph Giles committed
497
This would indicate that there are more samples to be skipped from the initial
498
 decoded output than exist in the stream.
499 500 501 502 503 504 505 506 507
If the granule position is smaller than the number of decoded samples produced
 by the packets that complete on that page, then a demuxer MUST use an initial
 granule position of '0', and can work forwards from '0' to timestamp
 individual packets.
If the granule position is larger than the number of decoded samples available,
 then the demuxer MUST still work backwards as described above, even if the
 'end of stream' flag is set, to determine the initial granule position, and
 thus the initial PCM sample position.
Both of these will be greater than '0' in this case.
508 509 510
</t>
</section>

511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
<section anchor="seeking_and_preroll" title="Seeking and Pre-roll">
<t>
Seeking in Ogg files is best performed using a bisection search for a page
 whose granule position corresponds to a PCM position at or before the seek
 target.
With appropriately weighted bisection, accurate seeking can be performed with
 just three or four bisections even in multi-gigabyte files.
See <xref target="seeking"/> for general implementation guidance.
</t>

<t>
When seeking within an Ogg Opus stream, the decoder SHOULD start decoding (and
 discarding the output) at least 3840&nbsp;samples (80&nbsp;ms) prior to the
 seek target in order to ensure that the output audio is correct by the time it
 reaches the seek target.
This 'pre-roll' is separate from, and unrelated to, the 'pre-skip' used at the
 beginning of the stream.
If the point 80&nbsp;ms prior to the seek target comes before the initial PCM
 sample position, the decoder SHOULD start decoding from the beginning of the
 stream, applying pre-skip as normal, regardless of whether the pre-skip is
Ralph Giles's avatar
Ralph Giles committed
531 532
 larger or smaller than 80&nbsp;ms, and then continue to discard samples
 to reach the seek target (if any).
533 534 535 536 537
</t>
</section>

</section>

538 539
<section anchor="headers" title="Header Packets">
<t>
540 541
An Opus stream contains exactly two mandatory header packets:
 an identification header and a comment header.
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
</t>

<section anchor="id_header" title="Identification Header">

<figure anchor="id_header_packet" title="ID Header Packet" align="center">
<artwork align="center"><![CDATA[
 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|      'O'      |      'p'      |      'u'      |      's'      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|      'H'      |      'e'      |      'a'      |      'd'      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|  Version = 1  | Channel Count |           Pre-skip            |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                     Input Sample Rate (Hz)                    |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|   Output Gain (Q7.8 in dB)    | Mapping Family|               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+               :
|                                                               |
:               Optional Channel Mapping Table...               :
|                                                               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
]]></artwork>
</figure>

<t>
The fields in the identification (ID) header have the following meaning:
<list style="numbers">
<t><spanx style="strong">Magic Signature</spanx>:
<vspace blankLines="1"/>
This is an 8-octet (64-bit) field that allows codec identification and is
 human-readable.
It contains, in order, the magic numbers:
<list style="empty">
<t>0x4F 'O'</t>
<t>0x70 'p'</t>
<t>0x75 'u'</t>
<t>0x73 's'</t>
<t>0x48 'H'</t>
<t>0x65 'e'</t>
<t>0x61 'a'</t>
<t>0x64 'd'</t>
</list>
Starting with "Op" helps distinguish it from audio data packets, as this is an
 invalid TOC sequence.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Version</spanx> (8 bits, unsigned):
<vspace blankLines="1"/>
The version number MUST always be '1' for this version of the encapsulation
 specification.
Implementations SHOULD treat streams where the upper four bits of the version
 number match that of a recognized specification as backwards-compatible with
 that specification.
That is, the version number can be split into "major" and "minor" version
 sub-fields, with changes to the "minor" sub-field (in the lower four bits)
 signaling compatible changes.
For example, a decoder implementing this specification SHOULD accept any stream
 with a version number of '15' or less, and SHOULD assume any stream with a
 version number '16' or greater is incompatible.
The initial version '1' was chosen to keep implementations from relying on this
 octet as a null terminator for the "OpusHead" string.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Output Channel Count</spanx> 'C' (8 bits, unsigned):
<vspace blankLines="1"/>
This is the number of output channels.
This might be different than the number of encoded channels, which can change
 on a packet-by-packet basis.
This value MUST NOT be zero.
The maximum allowable value depends on the channel mapping family, and might be
 as large as 255.
See <xref target="channel_mapping"/> for details.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Pre-skip</spanx> (16 bits, unsigned, little
 endian):
<vspace blankLines="1"/>
This is the number of samples (at 48&nbsp;kHz) to discard from the decoder
 output when starting playback, and also the number to subtract from a page's
 granule position to calculate its PCM sample position.
624 625 626
When cropping the beginning of existing Ogg Opus streams, a pre-skip of at
 least 3,840&nbsp;samples (80&nbsp;ms) is RECOMMENDED to ensure complete
 convergence in the decoder.
627 628 629 630 631 632 633 634
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Input Sample Rate</spanx> (32 bits, unsigned, little
 endian):
<vspace blankLines="1"/>
This field is <spanx style="emph">not</spanx> the sample rate to use for
 playback of the encoded data.
<vspace blankLines="1"/>
635 636
Opus can switch between internal audio bandwidths of 4, 6, 8, 12, and
 20&nbsp;kHz.
Ralph Giles's avatar
Ralph Giles committed
637
Each packet in the stream can have a different audio bandwidth.
638 639 640 641 642 643 644 645
Regardless of the audio bandwidth, the reference decoder supports decoding any
 stream at a sample rate of 8, 12, 16, 24, or 48&nbsp;kHz.
The original sample rate of the encoder input is not preserved by the lossy
 compression.
<vspace blankLines="1"/>
An Ogg Opus player SHOULD select the playback sample rate according to the
 following procedure:
<list style="numbers">
646 647 648 649
<t>If the hardware supports 48&nbsp;kHz playback, decode at 48&nbsp;kHz.</t>
<t>Otherwise, if the hardware's highest available sample rate is a supported
 rate, decode at this sample rate.</t>
<t>Otherwise, if the hardware's highest available sample rate is less than
Ralph Giles's avatar
Ralph Giles committed
650 651
 48&nbsp;kHz, decode at the next highest supported rate above this and
 resample.</t>
652
<t>Otherwise, decode at 48&nbsp;kHz and resample.</t>
653 654 655
</list>
However, the 'Input Sample Rate' field allows the encoder to pass the sample
 rate of the original input stream as metadata.
Ralph Giles's avatar
Ralph Giles committed
656
This is useful when the user requires the output sample rate to match the
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
 input sample rate.
For example, a non-player decoder writing PCM format samples to disk might
 choose to resample the output audio back to the original input sample rate to
 reduce surprise to the user, who might reasonably expect to get back a file
 with the same sample rate as the one they fed to the encoder.
<vspace blankLines="1"/>
A value of zero indicates 'unspecified'.
Encoders SHOULD write the actual input sample rate or zero, but decoder
 implementations which do something with this field SHOULD take care to behave
 sanely if given crazy values (e.g., do not actually upsample the output to
 10 MHz if requested).
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Output Gain</spanx> (16 bits, signed, little
 endian):
<vspace blankLines="1"/>
This is a gain to be applied by the decoder.
It is 20*log10 of the factor to scale the decoder output by to achieve the
 desired playback volume, stored in a 16-bit, signed, two's complement
 fixed-point value with 8 fractional bits (i.e., Q7.8).
<figure align="center">
Ralph Giles's avatar
Ralph Giles committed
678 679 680
<preamble>
To apply the gain, a decoder could use
</preamble>
681 682 683
<artwork align="center"><![CDATA[
sample *= pow(10, output_gain/(20.0*256)) ,
]]></artwork>
Ralph Giles's avatar
Ralph Giles committed
684
<postamble>
685
 where output_gain is the raw 16-bit value from the header.
Ralph Giles's avatar
Ralph Giles committed
686 687
</postamble>
</figure>
688
<vspace blankLines="1"/>
Ralph Giles's avatar
Ralph Giles committed
689
Virtually all players and media frameworks SHOULD apply it by default.
690
If a player chooses to apply any volume adjustment or gain modification, such
691 692 693
 as the R128_TRACK_GAIN (see <xref target="comment_header"/>), the adjustment
 MUST be applied in addition to this output gain in order to achieve playback
 at the normalized volume.
694 695 696
<vspace blankLines="1"/>
An encoder SHOULD set this field to zero, and instead apply any gain prior to
 encoding, when this is possible and does not conflict with the user's wishes.
Ralph Giles's avatar
Ralph Giles committed
697 698 699
A nonzero output gain indicates the gain was adjusted after encoding, or that
 a user wished to adjust the gain for playback while preserving the ability
 to recover the original signal amplitude.
700 701 702 703 704
<vspace blankLines="1"/>
Although the output gain has enormous range (+/- 128 dB, enough to amplify
 inaudible sounds to the threshold of physical pain), most applications can
 only reasonably use a small portion of this range around zero.
The large range serves in part to ensure that gain can always be losslessly
705
 transferred between OpusHead and R128 gain tags (see below) without
706 707 708 709 710 711
 saturating.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Channel Mapping Family</spanx> (8 bits,
 unsigned):
<vspace blankLines="1"/>
712
This octet indicates the order and semantic meaning of the output channels.
713 714 715 716 717 718
<vspace blankLines="1"/>
Each possible value of this octet indicates a mapping family, which defines a
 set of allowed channel counts, and the ordered set of channel names for each
 allowed channel count.
The details are described in <xref target="channel_mapping"/>.
</t>
719 720 721 722 723
<t><spanx style="strong">Channel Mapping Table</spanx>:
This table defines the mapping from encoded streams to output channels.
It is omitted when the channel mapping family is 0, but REQUIRED otherwise.
Its contents are specified in <xref target="channel_mapping"/>.
</t>
724 725 726
</list>
</t>

727 728 729 730 731 732 733 734 735 736 737 738 739 740
<t>
All fields in the ID headers are REQUIRED, except for the channel mapping
 table, which is omitted when the channel mapping family is 0.
Implementations SHOULD reject ID headers which do not contain enough data for
 these fields, even if they contain a valid Magic Signature.
Future versions of this specification, even backwards-compatible versions,
 might include additional fields in the ID header.
If an ID header has a compatible major version, but a larger minor version,
 an implementation MUST NOT reject it for containing additional data not
 specified here.
However, implementations MAY reject streams in which the ID header does not
 complete on the first page.
</t>

741 742 743 744 745 746
<section anchor="channel_mapping" title="Channel Mapping">
<t>
An Ogg Opus stream allows mapping one number of Opus streams (N) to a possibly
 larger number of decoded channels (M+N) to yet another number of output
 channels (C), which might be larger or smaller than the number of decoded
 channels.
Ralph Giles's avatar
Ralph Giles committed
747 748
The order and meaning of these channels are defined by a channel mapping,
 which consists of the 'channel mapping family' octet and, for channel mapping
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771
 families other than family&nbsp;0, a channel mapping table, as illustrated in
 <xref target="channel_mapping_table"/>.
</t>

<figure anchor="channel_mapping_table" title="Channel Mapping Table"
 align="center">
<artwork align="center"><![CDATA[
 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
                                                +-+-+-+-+-+-+-+-+
                                                | Stream Count  |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Coupled Count |              Channel Mapping...               :
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
]]></artwork>
</figure>

<t>
The fields in the channel mapping table have the following meaning:
<list style="numbers" counter="8">
<t><spanx style="strong">Stream Count</spanx> 'N' (8 bits, unsigned):
<vspace blankLines="1"/>
This is the total number of streams encoded in each Ogg packet.
Ralph Giles's avatar
Ralph Giles committed
772
This value is necessary to correctly parse the packed Opus packets inside an
773 774 775 776 777 778 779 780
 Ogg packet, as described in <xref target="packet_organization"/>.
This value MUST NOT be zero, as without at least one Opus packet with a valid
 TOC sequence, a demuxer cannot recover the duration of an Ogg packet.
<vspace blankLines="1"/>
For channel mapping family&nbsp;0, this value defaults to 1, and is not coded.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Coupled Stream Count</spanx> 'M' (8 bits, unsigned):
Ralph Giles's avatar
Ralph Giles committed
781
This is the number of streams whose decoders are to be configured to produce
782 783 784 785 786
 two channels.
This MUST be no larger than the total number of streams, N.
<vspace blankLines="1"/>
Each packet in an Opus stream has an internal channel count of 1 or 2, which
 can change from packet to packet.
787
This is selected by the encoder depending on the bitrate and the audio being
788 789 790 791 792 793 794 795
 encoded.
The original channel count of the encoder input is not preserved by the lossy
 compression.
<vspace blankLines="1"/>
Regardless of the internal channel count, any Opus stream can be decoded as
 mono (a single channel) or stereo (two channels) by appropriate initialization
 of the decoder.
The 'coupled stream count' field indicates that the first M Opus decoders are
796 797
 to be initialized for stereo output, and the remaining N-M decoders are to be
 initialized for mono only.
798 799 800 801 802 803 804 805 806
The total number of decoded channels, (M+N), MUST be no larger than 255, as
 there is no way to index more channels than that in the channel mapping.
<vspace blankLines="1"/>
For channel mapping family&nbsp;0, this value defaults to C-1 (i.e., 0 for mono
 and 1 for stereo), and is not coded.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Channel Mapping</spanx> (8*C bits):
This contains one octet per output channel, indicating which decoded channel
Ralph Giles's avatar
Ralph Giles committed
807
 is to be used for each one.
808 809 810 811 812
Let 'index' be the value of this octet for a particular output channel.
This value MUST either be smaller than (M+N), or be the special value 255.
If 'index' is less than 2*M, the output MUST be taken from decoding stream
 ('index'/2) as stereo and selecting the left channel if 'index' is even, and
 the right channel if 'index' is odd.
813 814
If 'index' is 2*M or larger, but less than 255, the output MUST be taken from
 decoding stream ('index'-M) as mono.
815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
If 'index' is 255, the corresponding output channel MUST contain pure silence.
<vspace blankLines="1"/>
The number of output channels, C, is not constrained to match the number of
 decoded channels (M+N).
A single index value MAY appear multiple times, i.e., the same decoded channel
 might be mapped to multiple output channels.
Some decoded channels might not be assigned to any output channel, as well.
<vspace blankLines="1"/>
For channel mapping family&nbsp;0, the first index defaults to 0, and if C==2,
 the second index defaults to 1.
Neither index is coded.
</t>
</list>
</t>

<t>
After producing the output channels, the channel mapping family determines the
 semantic meaning of each one.
Ralph Giles's avatar
Ralph Giles committed
833
There are three defined mapping families in this specification.
834 835 836 837
</t>

<section anchor="channel_mapping_0" title="Channel Mapping Family 0">
<t>
838
Allowed numbers of channels: 1 or 2.
839 840 841
RTP mapping.
</t>
<t>
842 843 844 845 846 847
<list style="symbols">
<t>1 channel: monophonic (mono).</t>
<t>2 channels: stereo (left, right).</t>
</list>
<spanx style="strong">Special mapping</spanx>: This channel mapping value also
 indicates that the contents consists of a single Opus stream that is stereo if
848 849 850
 and only if C==2, with stream index 0 mapped to output channel 0 (mono, or
 left channel) and stream index 1 mapped to output channel 1 (right channel)
 if stereo.
851 852 853
When the 'channel mapping family' octet has this value, the channel mapping
 table MUST be omitted from the ID header packet.
</t>
854 855 856 857
</section>

<section anchor="channel_mapping_1" title="Channel Mapping Family 1">
<t>
858
Allowed numbers of channels: 1...8.
859 860 861
Vorbis channel order.
</t>
<t>
862
Each channel is assigned to a speaker location in a conventional surround
863
 arrangement.
864 865 866 867 868 869 870
Specific locations depend on the number of channels, and are given below
 in order of the corresponding channel indicies.
<list style="symbols">
  <t>1 channel: monophonic (mono).</t>
  <t>2 channels: stereo (left, right).</t>
  <t>3 channels: linear surround (left, center, right)</t>
  <t>4 channels: quadraphonic (front&nbsp;left, front&nbsp;right, rear&nbsp;left, rear&nbsp;right).</t>
871 872 873 874
  <t>5 channels: 5.0 surround (front&nbsp;left, front&nbsp;center, front&nbsp;right, rear&nbsp;left, rear&nbsp;right).</t>
  <t>6 channels: 5.1 surround (front&nbsp;left, front&nbsp;center, front&nbsp;right, rear&nbsp;left, rear&nbsp;right, LFE).</t>
  <t>7 channels: 6.1 surround (front&nbsp;left, front&nbsp;center, front&nbsp;right, side&nbsp;left, side&nbsp;right, rear&nbsp;center, LFE).</t>
  <t>8 channels: 7.1 surround (front&nbsp;left, front&nbsp;center, front&nbsp;right, side&nbsp;left, side&nbsp;right, rear&nbsp;left, rear&nbsp;right, LFE)</t>
875
</list>
Ralph Giles's avatar
Ralph Giles committed
876 877
</t>
<t>
878 879
This set of surround options and speaker location orderings is the same
 as those used by the Vorbis codec <xref target="vorbis-mapping"/>.
880 881 882
The ordering is different from the one used by the
 WAVE <xref target="wave-multichannel"/> and
 FLAC <xref target="flac"/> formats,
Ralph Giles's avatar
Ralph Giles committed
883 884
 so correct ordering requires permutation of the output channels when decoding
 to or encoding from those formats.
885
'LFE' here refers to a Low Frequency Effects, often mapped to a subwoofer
Ralph Giles's avatar
Ralph Giles committed
886
 with no particular spatial position.
887
Implementations SHOULD identify 'side' or 'rear' speaker locations with
888 889
 'surround' and 'back' as appropriate when interfacing with audio formats
 or systems which prefer that terminology.
890
</t>
891 892 893 894 895 896 897 898 899
</section>

<section anchor="channel_mapping_255"
 title="Channel Mapping Family 255">
<t>
Allowed numbers of channels: 1...255.
No defined channel meaning.
</t>
<t>
900 901 902 903 904 905 906
Channels are unidentified.
General-purpose players SHOULD NOT attempt to play these streams, and offline
 decoders MAY deinterleave the output into separate PCM files, one per channel.
Decoders SHOULD NOT produce output for channels mapped to stream index 255
 (pure silence) unless they have no other way to indicate the index of
 non-silent channels.
</t>
907 908 909 910 911
</section>

<section anchor="channel_mapping_undefined"
 title="Undefined Channel Mappings">
<t>
912
The remaining channel mapping families (2...254) are reserved.
Ralph Giles's avatar
Ralph Giles committed
913
A decoder encountering a reserved channel mapping family value SHOULD act as
914
 though the value is 255.
915 916 917 918 919
</t>
</section>

<section anchor="downmix" title="Downmixing">
<t>
920 921 922 923 924 925 926
An Ogg Opus player MUST play any Ogg Opus stream with a channel mapping family
 of 0 or 1, even if the number of channels does not match the physically
 connected audio hardware.
Players SHOULD perform channel mixing to increase or reduce the number of
 channels as needed.
</t>

927 928 929 930 931 932
<t>
Implementations MAY use the following matricies to implement downmixing from
 multichannel files using <xref target="channel_mapping_1">Channel Mapping
 Family 1</xref>, which are known to give acceptable results for stereo.
Matricies for 3 and 4 channels are normalized so each coefficent row sums
 to 1 to avoid clipping.
Ralph Giles's avatar
Ralph Giles committed
933
For 5 or more channels they are normalized to 2 as a compromise between
934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949
 clipping and dynamic range reduction.
</t>
<t>
In these matricies the front left and front right channels are generally
passed through directly.
When a surround channel is split between both the left and right stereo
 channels, coefficients are chosen so their squares sum to 1, which
 helps preserve the perceived intensity.
Rear channels are mixed more diffusely or attenuated to maintain focus
 on the front channels.
</t>

<figure anchor="downmix-matrix-3"
 title="Stereo downmix matrix for the linear surround channel mapping"
 align="center">
<artwork align="center"><![CDATA[
950 951
L output = ( 0.585786 * left + 0.414214 * center                    )
R output = (                   0.414214 * center + 0.585786 * right )
952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
]]></artwork>
<postamble>
Exact coefficient values are 1 and 1/sqrt(2), multiplied by
 1/(1 + 1/sqrt(2)) for normalization.
</postamble>
</figure>

<figure anchor="downmix-matrix-4"
 title="Stereo downmix matrix for the quadraphonic channel mapping"
 align="center">
<artwork align="center"><![CDATA[
/          \   /                                     \ / FL \
| L output |   | 0.422650 0.000000 0.366025 0.211325 | | FR |
| R output | = | 0.000000 0.422650 0.211325 0.366025 | | RL |
\          /   \                                     / \ RR /
]]></artwork>
<postamble>
Exact coefficient values are 1, sqrt(3)/2 and 1/2, multiplied by
 1/(1&nbsp;+&nbsp;sqrt(3)/2&nbsp;+&nbsp;1/2) for normalization.
</postamble>
</figure>

<figure anchor="downmix-matrix-5"
 title="Stereo downmix matrix for the 5.0 surround mapping"
 align="center">
<artwork align="center"><![CDATA[
                                                         / FL \
/   \   /                                              \ | FC |
| L |   | 0.650802 0.460186 0.000000 0.563611 0.325401 | | FR |
| R | = | 0.000000 0.460186 0.650802 0.325401 0.563611 | | RL |
\   /   \                                              / | RR |
                                                         \    /
]]></artwork>
<postamble>
Exact coefficient values are 1, 1/sqrt(2), sqrt(3)/2 and 1/2, multiplied by
 2/(1&nbsp;+&nbsp;1/sqrt(2)&nbsp;+&nbsp;sqrt(3)/2&nbsp;+&nbsp;1/2)
 for normalization.
</postamble>
</figure>

<figure anchor="downmix-matrix-6"
 title="Stereo downmix matrix for the 5.1 surround mapping"
 align="center">
<artwork align="center"><![CDATA[
                                                                /FL \
/ \   /                                                       \ |FC |
|L|   | 0.529067 0.374107 0.000000 0.458186 0.264534 0.374107 | |FR |
|R| = | 0.000000 0.374107 0.529067 0.264534 0.458186 0.374107 | |RL |
\ /   \                                                       / |RR |
                                                                \LFE/
]]></artwork>
<postamble>
Exact coefficient values are 1, 1/sqrt(2), sqrt(3)/2 and 1/2, multiplied by
2/(1&nbsp;+&nbsp;1/sqrt(2)&nbsp;+&nbsp;sqrt(3)/2&nbsp;+&nbsp;1/2 + 1/sqrt(2))
 for normalization.
</postamble>
</figure>

<figure anchor="downmix-matrix-7"
 title="Stereo downmix matrix for the 6.1 surround mapping"
 align="center">
<artwork align="center"><![CDATA[
 /                                                                \
 | 0.455310 0.321953 0.000000 0.394310 0.227655 0.278819 0.321953 |
 | 0.000000 0.321953 0.455310 0.227655 0.394310 0.278819 0.321953 |
 \                                                                /
]]></artwork>
<postamble>
Exact coefficient values are 1, 1/sqrt(2), sqrt(3)/2, 1/2 and
 sqrt(3)/2/sqrt(2), multiplied by
 2/(1&nbsp;+&nbsp;1/sqrt(2)&nbsp;+&nbsp;sqrt(3)/2&nbsp;+&nbsp;1/2 +
 sqrt(3)/2/sqrt(2) + 1/sqrt(2)) for normalization.
The coeffients are in the same order as in <xref target="channel_mapping_1" />,
 and the matricies above.
</postamble>
</figure>

<figure anchor="downmix-matrix-8"
 title="Stereo downmix matrix for the 7.1 surround mapping"
 align="center">
<artwork align="center"><![CDATA[
/                                                                 \
| .388631 .274804 .000000 .336565 .194316 .336565 .194316 .274804 |
| .000000 .274804 .388631 .194316 .336565 .194316 .336565 .274804 |
\                                                                 /
]]></artwork>
<postamble>
Exact coefficient values are 1, 1/sqrt(2), sqrt(3)/2 and 1/2, multiplied by
 2/(2&nbsp;+&nbsp;2/sqrt(2)&nbsp;+&nbsp;sqrt(3)) for normalization.
The coeffients are in the same order as in <xref target="channel_mapping_1" />,
 and the matricies above.
</postamble>
</figure>
1045 1046 1047

</section>

1048 1049 1050 1051
</section> <!-- end channel_mapping_table -->

</section> <!-- end id_header -->

1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
<section anchor="comment_header" title="Comment Header">

<figure anchor="comment_header_packet" title="Comment Header Packet"
 align="center">
<artwork align="center"><![CDATA[
 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|      'O'      |      'p'      |      'u'      |      's'      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|      'T'      |      'a'      |      'g'      |      's'      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                     Vendor String Length                      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                                                               |
:                        Vendor String...                       :
|                                                               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                   User Comment List Length                    |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                 User Comment #0 String Length                 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                                                               |
:                   User Comment #0 String...                   :
|                                                               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                 User Comment #1 String Length                 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
:                                                               :
]]></artwork>
</figure>

<t>
The comment header consists of a 64-bit magic signature, followed by data in
 the same format as the <xref target="vorbis-comment"/> header used in Ogg
1087 1088
 Vorbis, except (like Ogg Theora and Speex) the final "framing bit" specified
 in the Vorbis spec is not present.
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
<list style="numbers">
<t><spanx style="strong">Magic Signature</spanx>:
<vspace blankLines="1"/>
This is an 8-octet (64-bit) field that allows codec identification and is
 human-readable.
It contains, in order, the magic numbers:
<list style="empty">
<t>0x4F 'O'</t>
<t>0x70 'p'</t>
<t>0x75 'u'</t>
<t>0x73 's'</t>
<t>0x54 'T'</t>
<t>0x61 'a'</t>
<t>0x67 'g'</t>
<t>0x73 's'</t>
</list>
Starting with "Op" helps distinguish it from audio data packets, as this is an
 invalid TOC sequence.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Vendor String Length</spanx> (32 bits, unsigned,
 little endian):
<vspace blankLines="1"/>
This field gives the length of the following vendor string, in octets.
It MUST NOT indicate that the vendor string is longer than the rest of the
 packet.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">Vendor String</spanx> (variable length, UTF-8 vector):
<vspace blankLines="1"/>
This is a simple human-readable tag for vendor information, encoded as a UTF-8
1120
 string&nbsp;<xref target="RFC3629"/>.
Ralph Giles's avatar
Ralph Giles committed
1121
No terminating null octet is necessary.
1122
<vspace blankLines="1"/>
Ralph Giles's avatar
Ralph Giles committed
1123
This tag is intended to identify the codec encoder and encapsulation
1124
 implementations, for tracing differences in technical behavior.
1125 1126
User-facing encoding applications can use the 'ENCODER' user comment tag
 to identify themselves.
Ralph Giles's avatar
Ralph Giles committed
1127
<vspace blankLines="1"/>
1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
</t>
<t><spanx style="strong">User Comment List Length</spanx> (32 bits, unsigned,
 little endian):
<vspace blankLines="1"/>
This field indicates the number of user-supplied comments.
It MAY indicate there are zero user-supplied comments, in which case there are
 no additional fields in the packet.
It MUST NOT indicate that there are so many comments that the comment string
 lengths would require more data than is available in the rest of the packet.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">User Comment #i String Length</spanx> (32 bits,
 unsigned, little endian):
<vspace blankLines="1"/>
This field gives the length of the following user comment string, in octets.
There is one for each user comment indicated by the 'user comment list length'
 field.
It MUST NOT indicate that the string is longer than the rest of the packet.
<vspace blankLines="1"/>
</t>
<t><spanx style="strong">User Comment #i String</spanx> (variable length, UTF-8
 vector):
<vspace blankLines="1"/>
This field contains a single user comment string.
There is one for each user comment indicated by the 'user comment list length'
 field.
</t>
</list>
</t>

1158 1159 1160 1161 1162 1163
<t>
The vendor string length and user comment list length are REQUIRED, and
 implementations SHOULD reject comment headers that do not contain enough data
 for these fields, or that do not contain enough data for the corresponding
 vendor string or user comments they describe.
Making this check before allocating the associated memory to contain the data
Ralph Giles's avatar
Ralph Giles committed
1164
 helps prevent a possible Denial-of-Service (DoS) attack from small comment
1165 1166 1167 1168
 headers that claim to contain strings longer than the entire packet or more
 user comments than than could possibly fit in the packet.
</t>

1169 1170 1171 1172
<t>
Immediately following the user comment list, the comment header MAY
 contain zero-padding or other binary data which is not specified here.
If the least-significant bit of the first byte of this data is 1, then editors
Ralph Giles's avatar
Ralph Giles committed
1173 1174 1175
 SHOULD preserve the contents of this data when updating the tags, but if this
 bit is 0, all such data MAY be treated as padding, and truncated or discarded
 as desired.
1176 1177
</t>

1178
<section anchor="comment_format" title="Tag Definitions">
1179 1180
<t>
The user comment strings follow the NAME=value format described by
1181 1182
 <xref target="vorbis-comment"/> with the same recommended tag names:
 ARTIST, TITLE, DATE, ALBUM, and so on.
Ralph Giles's avatar
Ralph Giles committed
1183
</t>
1184
<t>
1185
Two new comment tags are introduced here:
1186 1187
</t>

1188
<figure align="center">
1189
  <preamble>An optional gain for track nomalization</preamble>
1190 1191 1192
<artwork align="left"><![CDATA[
R128_TRACK_GAIN=-573
]]></artwork>
Ralph Giles's avatar
Ralph Giles committed
1193
<postamble>
1194 1195
representing the volume shift needed to normalize the track's volume
 during isolated playback, in random shuffle, and so on.
1196 1197
The gain is a Q7.8 fixed point number in dB, as in the ID header's 'output
 gain' field.
Ralph Giles's avatar
Ralph Giles committed
1198 1199 1200
</postamble>
</figure>
<t>
1201 1202 1203 1204
This tag is similar to the REPLAYGAIN_TRACK_GAIN tag in
 Vorbis&nbsp;<xref target="replay-gain"/>, except that the normal volume
 reference is the <xref target="EBU-R128"/> standard.
</t>
1205 1206
<figure align="center">
  <preamble>An optional gain for album nomalization</preamble>
1207 1208 1209 1210
<artwork align="left"><![CDATA[
R128_ALBUM_GAIN=111
]]></artwork>
<postamble>
1211
representing the volume shift needed to normalize the overall volume when
1212
 played as part of a particular collection of tracks.
1213 1214
The gain is also a Q7.8 fixed point number in dB, as in the ID header's
 'output gain' field.
1215 1216
</postamble>
</figure>
1217
<t>
1218
An Ogg Opus stream MUST NOT have more than one of each tag, and if present
1219
 their values MUST be an integer from -32768 to 32767, inclusive,
1220
 represented in ASCII as a base 10 number with no whitespace.
1221
A leading '+' or '-' character is valid.
Ralph Giles's avatar
Ralph Giles committed
1222
Leading zeros are also permitted, but the value MUST be represented by
1223
 no more than 6 characters.
1224 1225 1226
Other non-digit characters MUST NOT be present.
</t>
<t>
1227 1228 1229
If present, R128_TRACK_GAIN and R128_ALBUM_GAIN MUST correctly represent
 the R128 normalization gain relative to the 'output gain' field specified
 in the ID header.
1230
If a player chooses to make use of the R128_TRACK_GAIN tag or the
1231 1232
 R128_ALBUM_GAIN tag, it MUST apply those gains
 <spanx style="emph">in addition</spanx> to the 'output gain' value.
1233
If a tool modifies the ID header's 'output gain' field, it MUST also update or
1234
 remove the R128_TRACK_GAIN and R128_ALBUM_GAIN comment tags if present.
Ralph Giles's avatar
Ralph Giles committed
1235
An encoder SHOULD assume that by default tools will respect the 'output gain'
1236
 field, and not the comment tag.
1237 1238 1239 1240 1241
</t>
<t>
To avoid confusion with multiple normalization schemes, an Opus comment header
 SHOULD NOT contain any of the REPLAYGAIN_TRACK_GAIN, REPLAYGAIN_TRACK_PEAK,
 REPLAYGAIN_ALBUM_GAIN, or REPLAYGAIN_ALBUM_PEAK tags.
1242
<xref target="EBU-R128"/> normalization is preferred to the earlier
1243
 REPLAYGAIN schemes because of its clear definition and adoption by industry.
Ralph Giles's avatar
Ralph Giles committed
1244
Peak normalizations are difficult to calculate reliably for lossy codecs
1245
 because of variation in excursion heights due to decoder differences.
1246 1247
In the authors' investigations they were not applied consistently or broadly
 enough to merit inclusion here.
1248
</t>
1249 1250
</section> <!-- end comment_format -->
</section> <!-- end comment_header -->
1251

1252
</section> <!-- end headers -->
1253

1254
<section anchor="packet_size_limits" title="Packet Size Limits">
1255
<t>
1256
Technically, valid Opus packets can be arbitrarily large due to the padding
1257 1258
 format, although the amount of non-padding data they can contain is bounded.
These packets might be spread over a similarly enormous number of Ogg pages.
Ralph Giles's avatar
Ralph Giles committed
1259 1260
Encoders SHOULD use no more padding than is necessary to make a variable
 bitrate (VBR) stream constant bitrate (CBR).
1261 1262 1263
Decoders SHOULD avoid attempting to allocate excessive amounts of memory when
 presented with a very large packet.
The presence of an extremely large packet in the stream could indicate a
Ralph Giles's avatar
Ralph Giles committed
1264
 memory exhaustion attack or stream corruption.
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
Decoders SHOULD reject a packet that is too large to process, and display a
 warning message.
</t>
<t>
In an Ogg Opus stream, the largest possible valid packet that does not use
 padding has a size of (61,298*N&nbsp;-&nbsp;2) octets, or about 60&nbsp;kB per
 Opus stream.
With 255&nbsp;streams, this is 15,630,988&nbsp;octets (14.9&nbsp;MB) and can
 span up to 61,298&nbsp;Ogg pages, all but one of which will have a granule
 position of -1.
This is of course a very extreme packet, consisting of 255&nbsp;streams, each
 containing 120&nbsp;ms of audio encoded as 2.5&nbsp;ms frames, each frame
 using the maximum possible number of octets (1275) and stored in the least
 efficient manner allowed (a VBR code&nbsp;3 Opus packet).
Ralph Giles's avatar
Ralph Giles committed
1279 1280
Even in such a packet, most of the data will be zeros as 2.5&nbsp;ms frames
 cannot actually use all 1275&nbsp;octets.
1281 1282 1283
The largest packet consisting of entirely useful data is
 (15,326*N&nbsp;-&nbsp;2) octets, or about 15&nbsp;kB per stream.
This corresponds to 120&nbsp;ms of audio encoded as 10&nbsp;ms frames in either
1284
 SILK or Hybrid mode, but at a data rate of over 1&nbsp;Mbps, which makes little
1285 1286 1287
 sense for the quality achieved.
A more reasonable limit is (7,664*N&nbsp;-&nbsp;2) octets, or about 7.5&nbsp;kB
 per stream.
1288
This corresponds to 120&nbsp;ms of audio encoded as 20&nbsp;ms stereo CELT mode
1289 1290
 frames, with a total bitrate just under 511&nbsp;kbps (not counting the Ogg
 encapsulation overhead).
1291
With N=8, the maximum number of channels currently defined by mapping
1292 1293 1294 1295 1296 1297 1298 1299 1300
 family&nbsp;1, this gives a maximum packet size of 61,310&nbsp;octets, or just
 under 60&nbsp;kB.
This is still quite conservative, as it assumes each output channel is taken
 from one decoded channel of a stereo packet.
An implementation could reasonably choose any of these numbers for its internal
 limits.
</t>
</section>

Ralph Giles's avatar
Ralph Giles committed
1301 1302
<section anchor="encoder" title="Encoder Guidelines">
<t>
Ralph Giles's avatar
Ralph Giles committed
1303
When encoding Opus streams, Ogg encoders SHOULD take into account the
Ralph Giles's avatar
Ralph Giles committed
1304
 algorithmic delay of the Opus encoder.
1305 1306 1307
</t>
<figure align="center">
<preamble>
Ralph Giles's avatar
Ralph Giles committed
1308 1309
In encoders derived from the reference implementation, the number of
 samples can be queried with:
1310 1311
</preamble>
<artwork align="center"><![CDATA[
1312
 opus_encoder_ctl(encoder_state, OPUS_GET_LOOKAHEAD, &delay_samples);
1313 1314 1315
]]></artwork>
</figure>
<t>
Ralph Giles's avatar
Ralph Giles committed
1316
To achieve good quality in the very first samples of a stream, the Ogg encoder
1317 1318 1319 1320 1321
 MAY use linear predictive coding (LPC) extrapolation
 <xref target="linear-prediction"/> to generate at least 120 extra samples at
 the beginning to avoid the Opus encoder having to encode a discontinuous
 signal.
For an input file containing 'length' samples, the Ogg encoder SHOULD set the
1322
 pre-skip header value to delay_samples+extra_samples, encode at least
1323 1324
 length+delay_samples+extra_samples samples, and set the granulepos of the last
 page to length+delay_samples+extra_samples.
Ralph Giles's avatar
Ralph Giles committed
1325 1326 1327 1328 1329 1330 1331 1332
This ensures that the encoded file has the same duration as the original, with
 no time offset. The best way to pad the end of the stream is to also use LPC
 extrapolation, but zero-padding is also acceptable.
</t>

<section anchor="lpc" title="LPC Extrapolation">
<t>
The first step in LPC extrapolation is to compute linear prediction
1333
 coefficients. <xref target="lpc-sample"/>
Ralph Giles's avatar
Ralph Giles committed
1334 1335 1336 1337
When extending the end of the signal, order-N (typically with N ranging from 8
 to 40) LPC analysis is performed on a window near the end of the signal.
The last N samples are used as memory to an infinite impulse response (IIR)
 filter.
Ralph Giles's avatar
Ralph Giles committed
1338 1339 1340
</t>
<figure align="center">
<preamble>
Ralph Giles's avatar
Ralph Giles committed
1341 1342 1343
The filter is then applied on a zero input to extrapolate the end of the signal.
Let a(k) be the kth LPC coefficient and x(n) be the nth sample of the signal,
 each new sample past the end of the signal is computed as:
Ralph Giles's avatar
Ralph Giles committed
1344
</preamble>
Ralph Giles's avatar
Ralph Giles committed
1345 1346 1347 1348 1349 1350 1351 1352
<artwork align="center"><![CDATA[
        N
       ---
x(n) = \   a(k)*x(n-k)
       /
       ---
       k=1
]]></artwork>
Ralph Giles's avatar
Ralph Giles committed
1353 1354
</figure>
<t>
Ralph Giles's avatar
Ralph Giles committed
1355 1356 1357 1358
The process is repeated independently for each channel.
It is possible to extend the beginning of the signal by applying the same
 process backward in time.
When extending the beginning of the signal, it is best to apply a "fade in" to
Ralph Giles's avatar
Ralph Giles committed
1359 1360
 the extrapolated signal, e.g. by multiplying it by a half-Hanning window
 <xref target="hanning"/>.
Ralph Giles's avatar
Ralph Giles committed
1361 1362 1363 1364 1365 1366 1367
</t>

</section>

<section anchor="continuous_chaining" title="Continuous Chaining">
<t>
In some applications, such as Internet radio, it is desirable to cut a long
Ron's avatar
Ron committed
1368
 stream into smaller chains, e.g. so the comment header can be updated.
Ralph Giles's avatar
Ralph Giles committed
1369 1370 1371 1372 1373 1374 1375
This can be done simply by separating the input streams into segments and
 encoding each segment independently.
The drawback of this approach is that it creates a small discontinuity
 at the boundary due to the lossy nature of Opus.
An encoder MAY avoid this discontinuity by using the following procedure:
<list style="numbers">
<t>Encode the last frame of the first segment as an independent frame by
1376 1377
 turning off all forms of inter-frame prediction.
De-emphasis is allowed.</t>
Ralph Giles's avatar
Ralph Giles committed
1378 1379 1380 1381
<t>Set the granulepos of the last page to a point near the end of the last
 frame.</t>
<t>Begin the second segment with a copy of the last frame of the first
 segment.</t>
1382
<t>Set the pre-skip value of the second stream in such a way as to properly
Ralph Giles's avatar
Ralph Giles committed
1383 1384 1385 1386 1387
 join the two streams.</t>
<t>Continue the encoding process normally from there, without any reset to
 the encoder.</t>
</list>
</t>
1388 1389 1390 1391 1392 1393 1394 1395 1396
<figure align="center">
<preamble>
In encoders derived from the reference implementation, inter-frame prediction
 can be turned off by calling:
</preamble>
<artwork align="center"><![CDATA[
 opus_encoder_ctl(encoder_state, OPUS_SET_PREDICTION_DISABLED, 1);
]]></artwork>
<postamble>
Ralph Giles's avatar
Ralph Giles committed
1397 1398
For best results, this implementation requires that prediction be explicitly
 enabled again before resuming normal encoding, even after a reset.
1399 1400 1401
</postamble>
</figure>

Ralph Giles's avatar
Ralph Giles committed
1402 1403
</section>

Ralph Giles's avatar
Ralph Giles committed
1404 1405
</section>

1406 1407
<section anchor="implementation" title="Implementation Status">
<t>
1408 1409 1410
A brief summary of major implementations of this draft is available
 at <eref target="https://wiki.xiph.org/OggOpusImplementation"/>,
  along with their status.
1411 1412
</t>
<t>
1413
[Note to RFC Editor: please remove this entire section before
1414
 final publication per <xref target="RFC6982"/>.]
1415 1416 1417
</t>
</section>

1418 1419 1420 1421 1422 1423 1424
<section anchor="security" title="Security Considerations">
<t>
Implementations of the Opus codec need to take appropriate security
 considerations into account, as outlined in <xref target="RFC4732"/>.
This is just as much a problem for the container as it is for the codec itself.
It is extremely important for the decoder to be robust against malicious
 payloads.
Ralph Giles's avatar
Ralph Giles committed