1 <?xml version="1.0" standalone="no"?>
2 <!DOCTYPE section PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
3 "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd" [
7 <section id="vorbis-spec-intro">
10 $Id: 01-introduction.xml,v 1.5 2002/10/27 14:55:31 giles Exp $
11 <emphasis>Last update to this document: July 18, 2002</emphasis>
14 <title>introduction and description</title>
17 <title>Overview</title>
20 This document provides a high level description of the Vorbis codec's
21 construction. A bit-by-bit specification appears beginning in <xref linkend="vorbis-spec-codec"/>.
22 The later sections assume a high-level
23 understanding of the Vorbis decode process, which is
27 <title>Application</title>
29 Vorbis is a general purpose perceptual audio CODEC intended to allow
30 maximum encoder flexibility, thus allowing it to scale competitively
31 over an exceptionally wide range of bitrates. At the high
32 quality/bitrate end of the scale (CD or DAT rate stereo, 16/24 bits)
33 it is in the same league as MPEG-2 and MPC. Similarly, the 1.0
34 encoder can encode high-quality CD and DAT rate stereo at below 48kbps
35 without resampling to a lower rate. Vorbis is also intended for
36 lower and higher sample rates (from 8kHz telephony to 192kHz digital
37 masters) and a range of channel representations (monaural,
38 polyphonic, stereo, quadraphonic, 5.1, ambisonic, or up to 255
44 <title>Classification</title>
46 Vorbis I is a forward-adaptive monolithic transform CODEC based on the
47 Modified Discrete Cosine Transform. The codec is structured to allow
48 addition of a hybrid wavelet filterbank in Vorbis II to offer better
49 transient response and reproduction using a transform better suited to
50 localized time events.
55 <title>Assumptions</title>
58 The Vorbis CODEC design assumes a complex, psychoacoustically-aware
59 encoder and simple, low-complexity decoder. Vorbis decode is
60 computationally simpler than mp3, although it does require more
61 working memory as Vorbis has no static probability model; the vector
62 codebooks used in the first stage of decoding from the bitstream are
63 packed in their entirety into the Vorbis bitstream headers. In
64 packed form, these codebooks occupy only a few kilobytes; the extent
65 to which they are pre-decoded into a cache is the dominant factor in
70 Vorbis provides none of its own framing, synchronization or protection
71 against errors; it is solely a method of accepting input audio,
72 dividing it into individual frames and compressing these frames into
73 raw, unformatted 'packets'. The decoder then accepts these raw
74 packets in sequence, decodes them, synthesizes audio frames from
75 them, and reassembles the frames into a facsimile of the original
76 audio stream. Vorbis is a free-form variable bit rate (VBR) codec and packets have no
77 minimum size, maximum size, or fixed/expected size. Packets
78 are designed that they may be truncated (or padded) and remain
79 decodable; this is not to be considered an error condition and is used
80 extensively in bitrate management in peeling. Both the transport
81 mechanism and decoder must allow that a packet may be any size, or
82 end before or after packet decode expects.</para>
85 Vorbis packets are thus intended to be used with a transport mechanism
86 that provides free-form framing, sync, positioning and error correction
87 in accordance with these design assumptions, such as Ogg (for file
88 transport) or RTP (for network multicast). For purposes of a few
89 examples in this document, we will assume that Vorbis is to be
90 embedded in an Ogg stream specifically, although this is by no means a
91 requirement or fundamental assumption in the Vorbis design.</para>
94 The specification for embedding Vorbis into
95 an Ogg transport stream is in a <xref linkend="vorbis-over-ogg"/>.
101 <title>Codec Setup and Probability Model</title>
104 Vorbis' heritage is as a research CODEC and its current design
105 reflects a desire to allow multiple decades of continuous encoder
106 improvement before running out of room within the codec specification.
107 For these reasons, configurable aspects of codec setup intentionally
108 lean toward the extreme of forward adaptive.</para>
111 The single most controversial design decision in Vorbis (and the most
112 unusual for a Vorbis developer to keep in mind) is that the entire
113 probability model of the codec, the Huffman and VQ codebooks, is
114 packed into the bitstream header along with extensive CODEC setup
115 parameters (often several hundred fields). This makes it impossible,
116 as it would be with MPEG audio layers, to embed a simple frame type
117 flag in each audio packet, or begin decode at any frame in the stream
118 without having previously fetched the codec setup header.
122 Vorbis <emphasis>can</emphasis> initiate decode at any arbitrary packet within a
123 bitstream so long as the codec has been initialized/setup with the
124 setup headers.</para></note>
127 Thus, Vorbis headers are both required for decode to begin and
128 relatively large as bitstream headers go. The header size is
129 unbounded, although for streaming a rule-of-thumb of 4kB or less is
130 recommended (and Xiph.Org's Vorbis encoder follows this suggestion).</para>
133 Our own design work indicates the the primary liability of the
134 required header is in mindshare; it is an unusual design and thus
135 causes some amount of complaint among engineers as this runs against
136 current design trends (and also points out limitations in some
137 existing software/interface designs, such as Windows' ACM codec
138 framework). However, we find that it does not fundamentally limit
139 Vorbis' suitable application space.</para>
143 <section><title>Format Specification</title>
145 The Vorbis format is well-defined by its decode specification; any
146 encoder that produces packets that are correctly decoded by the
147 reference Vorbis decoder described below may be considered a proper
148 Vorbis encoder. A decoder must faithfully and completely implement
149 the specification defined below (except where noted) to be considered
150 a proper Vorbis decoder.</para>
153 <section><title>Hardware Profile</title>
155 Although Vorbis decode is computationally simple, it may still run
156 into specific limitations of an embedded design. For this reason,
157 embedded designs are allowed to deviate in limited ways from the
158 'full' decode specification yet still be certified compliant. These
159 optional omissions are labelled in the spec where relevant.</para>
165 <title>Decoder Configuration</title>
168 Decoder setup consists of configuration of multiple, self-contained
169 component abstractions that perform specific functions in the decode
170 pipeline. Each different component instance of a specific type is
171 semantically interchangeable; decoder configuration consists both of
172 internal component configuration, as well as arrangement of specific
173 instances into a decode pipeline. Componentry arrangement is roughly
178 <imagedata fileref="components.png" format="PNG"/>
181 <phrase>decoder pipeline configuration</phrase>
185 <section><title>Global Config</title>
187 Global codec configuration consists of a few audio related fields
188 (sample rate, channels), Vorbis version (always '0' in Vorbis I),
189 bitrate hints, and the lists of component instances. All other
190 configuration is in the context of specific components.</para>
193 <section><title>Mode</title>
196 Each Vorbis frame is coded according to a master 'mode'. A bitstream
197 may use one or many modes.</para>
200 The mode mechanism is used to encode a frame according to one of
201 multiple possible methods with the intention of choosing a method best
202 suited to that frame. Different modes are, e.g. how frame size
203 is changed from frame to frame. The mode number of a frame serves as a
204 top level configuration switch for all other specific aspects of frame
208 A 'mode' configuration consists of a frame size setting, window type
209 (always 0, the Vorbis window, in Vorbis I), transform type (always
210 type 0, the MDCT, in Vorbis I) and a mapping number. The mapping
211 number specifies which mapping configuration instance to use for
212 low-level packet decode and synthesis.</para>
216 <section><title>Mapping</title>
219 A mapping contains a channel coupling description and a list of
220 'submaps' that bundle sets of channel vectors together for grouped
221 encoding and decoding. These submaps are not references to external
222 components; the submap list is internal and specific to a mapping.</para>
225 A 'submap' is a configuration/grouping that applies to a subset of
226 floor and residue vectors within a mapping. The submap functions as a
227 last layer of indirection such that specific special floor or residue
228 settings can be applied not only to all the vectors in a given mode,
229 but also specific vectors in a specific mode. Each submap specifies
230 the proper floor and residue instance number to use for decoding that
231 submap's spectral floor and spectral residue vectors.</para>
234 As an example:</para>
237 Assume a Vorbis stream that contains six channels in the standard 5.1
238 format. The sixth channel, as is normal in 5.1, is bass only.
239 Therefore it would be wasteful to encode a full-spectrum version of it
240 as with the other channels. The submapping mechanism can be used to
241 apply a full range floor and residue encoding to channels 0 through 4,
242 and a bass-only representation to the bass channel, thus saving space.
243 In this example, channels 0-4 belong to submap 0 (which indicates use
244 of a full-range floor) and channel 5 belongs to submap 1, which uses a
245 bass-only representation.</para>
249 <section><title>Floor</title>
252 Vorbis encodes a spectral 'floor' vector for each PCM channel. This
253 vector is a low-resolution representation of the audio spectrum for
254 the given channel in the current frame, generally used akin to a
255 whitening filter. It is named a 'floor' because the Xiph.Org
256 reference encoder has historically used it as a unit-baseline for
257 spectral resolution.</para>
260 A floor encoding may be of two types. Floor 0 uses a packed LSP
261 representation on a dB amplitude scale and Bark frequency scale.
262 Floor 1 represents the curve as a piecewise linear interpolated
263 representation on a dB amplitude scale and linear frequency scale.
264 The two floors are semantically interchangeable in
265 encoding/decoding. However, floor type 1 provides more stable
266 inter-frame behavior, and so is the preferred choice in all
267 coupled-stereo and high bitrate modes. Floor 1 is also considerably
268 less expensive to decode than floor 0.</para>
271 Floor 0 is not to be considered deprecated, but it is of limited
272 modern use. No known Vorbis encoder past Xiph.org's own beta 4 makes
273 use of floor 0.</para>
276 The values coded/decoded by a floor are both compactly formatted and
277 make use of entropy coding to save space. For this reason, a floor
278 configuration generally refers to multiple codebooks in the codebook
279 component list. Entropy coding is thus provided as an abstraction,
280 and each floor instance may choose from any and all available
281 codebooks when coding/decoding.</para>
285 <section><title>Residue</title>
287 The spectral residue is the fine structure of the audio spectrum
288 once the floor curve has been subtracted out. In simplest terms, it
289 is coded in the bitstream using cascaded (multi-pass) vector
290 quantization according to one of three specific packing/coding
291 algorithms numbered 0 through 2. The packing algorithm details are
292 configured by residue instance. As with the floor components, the
293 final VQ/entropy encoding is provided by external codebook instances
294 and each residue instance may choose from any and all available
298 <section><title>Codebooks</title>
301 Codebooks are a self-contained abstraction that perform entropy
302 decoding and, optionally, use the entropy-decoded integer value as an
303 offset into an index of output value vectors, returning the indicated
304 vector of values.</para>
307 The entropy coding in a Vorbis I codebook is provided by a standard
308 Huffman binary tree representation. This tree is tightly packed using
309 one of several methods, depending on whether codeword lengths are
310 ordered or unordered, or the tree is sparse.</para>
313 The codebook vector index is similarly packed according to index
314 characteristic. Most commonly, the vector index is encoded as a
315 single list of values of possible values that are then permuted into
316 a list of n-dimensional rows (lattice VQ).</para>
324 <title>High-level Decode Process</title>
327 <title>Decode setup</title>
330 Before decoding can begin, a decoder must initialize using the
331 bitstream headers matching the stream to be decoded. Vorbis uses
332 three header packets; all are required, in-order, by this
333 specification. Once set up, decode may begin at any audio packet
334 belonging to the Vorbis stream. In Vorbis I, all packets after the
335 three initial headers are audio packets. </para>
338 The header packets are, in order, the identification
339 header, the comments header, and the setup header.</para>
341 <section><title>Identification Header</title>
343 The identification header identifies the bitstream as Vorbis, Vorbis
344 version, and the simple audio characteristics of the stream such as
345 sample rate and number of channels.</para>
348 <section><title>Comment Header</title>
350 The comment header includes user text comments ("tags") and a vendor
351 string for the application/library that produced the bitstream. The
352 encoding and proper use of the comment header is described in
353 <xref linkend="vorbis-spec-comment"/>.</para>
356 <section><title>Setup Header</title>
358 The setup header includes extensive CODEC setup information as well as
359 the complete VQ and Huffman codebooks needed for decode.</para>
364 <section><title>Decode Procedure</title>
368 The decoding and synthesis procedure for all audio packets is
369 fundamentally the same.
371 <listitem><simpara>decode packet type flag</simpara></listitem>
372 <listitem><simpara>decode mode number</simpara></listitem>
373 <listitem><simpara>decode window shape (long windows only)</simpara></listitem>
374 <listitem><simpara>decode floor</simpara></listitem>
375 <listitem><simpara>decode residue into residue vectors</simpara></listitem>
376 <listitem><simpara>inverse channel coupling of residue vectors</simpara></listitem>
377 <listitem><simpara>generate floor curve from decoded floor data</simpara></listitem>
378 <listitem><simpara>compute dot product of floor and residue, producing audio spectrum vector</simpara></listitem>
379 <listitem><simpara>inverse monolithic transform of audio spectrum vector, always an MDCT in Vorbis I</simpara></listitem>
380 <listitem><simpara>overlap/add left-hand output of transform with right-hand output of previous frame</simpara></listitem>
381 <listitem><simpara>store right hand-data from transform of current frame for future lapping</simpara></listitem>
382 <listitem><simpara>if not first frame, return results of overlap/add as audio result of current frame</simpara></listitem>
388 Note that clever rearrangement of the synthesis arithmetic is
389 possible; as an example, one can take advantage of symmetries in the
390 MDCT to store the right-hand transform data of a partial MDCT for a
391 50% inter-frame buffer space savings, and then complete the transform
392 later before overlap/add with the next frame. This optimization
393 produces entirely equivalent output and is naturally perfectly legal.
394 The decoder must be <emphasis>entirely mathematically equivalent</emphasis> to the
395 specification, it need not be a literal semantic implementation.</para>
397 <section><title>Packet type decode</title>
400 Vorbis I uses four packet types. The first three packet types mark each
401 of the three Vorbis headers described above. The fourth packet type
402 marks an audio packet. All other packet types are reserved; packets
403 marked with a reserved type should be ignored.</para>
406 Following the three header packets, all packets in a Vorbis I stream
407 are audio. The first step of audio packet decode is to read and
408 verify the packet type; <emphasis>a non-audio packet when audio is expected
409 indicates stream corruption or a non-compliant stream. The decoder
410 must ignore the packet and not attempt decoding it to
411 audio</emphasis>.</para>
416 <section><title>Mode decode</title>
418 Vorbis allows an encoder to set up multiple, numbered packet 'modes',
419 as described earlier, all of which may be used in a given Vorbis
420 stream. The mode is encoded as an integer used as a direct offset into
421 the mode instance index. </para>
424 <section id="vorbis-spec-window">
425 <title>Window shape decode (long windows only)</title>
428 Vorbis frames may be one of two PCM sample sizes specified during
429 codec setup. In Vorbis I, legal frame sizes are powers of two from 64
430 to 8192 samples. Aside from coupling, Vorbis handles channels as
431 independent vectors and these frame sizes are in samples per channel.</para>
434 Vorbis uses an overlapping transform, namely the MDCT, to blend one
435 frame into the next, avoiding most inter-frame block boundary
436 artifacts. The MDCT output of one frame is windowed according to MDCT
437 requirements, overlapped 50% with the output of the previous frame and
438 added. The window shape assures seamless reconstruction. </para>
441 This is easy to visualize in the case of equal sized-windows:</para>
445 <imagedata fileref="window1.png" format="PNG"/>
448 <phrase>overlap of two equal-sized windows</phrase>
453 And slightly more complex in the case of overlapping unequal sized
458 <imagedata fileref="window2.png" format="PNG"/>
461 <phrase>overlap of a long and a short window</phrase>
466 In the unequal-sized window case, the window shape of the long window
467 must be modified for seamless lapping as above. It is possible to
468 correctly infer window shape to be applied to the current window from
469 knowing the sizes of the current, previous and next window. It is
470 legal for a decoder to use this method. However, in the case of a long
471 window (short windows require no modification), Vorbis also codes two
472 flag bits to specify pre- and post- window shape. Although not
473 strictly necessary for function, this minor redundancy allows a packet
474 to be fully decoded to the point of lapping entirely independently of
475 any other packet, allowing easier abstraction of decode layers as well
476 as allowing a greater level of easy parallelism in encode and
480 A description of valid window functions for use with an inverse MDCT
481 can be found in the paper
482 <citetitle pubwork="article">
483 <ulink url="http://www.iocon.com/resource/docs/ps/eusipco_corrected.ps">
484 The use of multirate filter banks for coding of high quality digital
485 audio</ulink></citetitle>, by T. Sporer, K. Brandenburg and B. Edler. Vorbis windows
486 all use the slope function
488 <!-- this equation is incorrect -->
489 <alt>y=sin(2PI*sin^2(x/n))</alt>
492 <phrase>$y = \sin(2\pi \, \sin^2(x/n))$</phrase>
500 <section><title>floor decode</title>
502 Each floor is encoded/decoded in channel order, however each floor
503 belongs to a 'submap' that specifies which floor configuration to
504 use. All floors are decoded before residue decode begins.</para>
507 <section><title>residue decode</title>
510 Although the number of residue vectors equals the number of channels,
511 channel coupling may mean that the raw residue vectors extracted
512 during decode do not map directly to specific channels. When channel
513 coupling is in use, some vectors will correspond to coupled magnitude
514 or angle. The coupling relationships are described in the codec setup
515 and may differ from frame to frame, due to different mode numbers.</para>
518 Vorbis codes residue vectors in groups by submap; the coding is done
519 in submap order from submap 0 through n-1. This differs from floors
520 which are coded using a configuration provided by submap number, but
521 are coded individually in channel order.</para>
525 <section><title>inverse channel coupling</title>
528 A detailed discussion of stereo in the Vorbis codec can be found in
529 the document <ulink url="stereo.html"><citetitle>Stereo Channel Coupling in the
530 Vorbis CODEC</citetitle></ulink>. Vorbis is not limited to only stereo coupling, but
531 the stereo document also gives a good overview of the generic coupling
535 Vorbis coupling applies to pairs of residue vectors at a time;
536 decoupling is done in-place a pair at a time in the order and using
537 the vectors specified in the current mapping configuration. The
538 decoupling operation is the same for all pairs, converting square
539 polar representation (where one vector is magnitude and the second
540 angle) back to Cartesian representation.</para>
543 After decoupling, in order, each pair of vectors on the coupling list,
544 the resulting residue vectors represent the fine spectral detail
545 of each output channel.</para>
549 <section><title>generate floor curve</title>
552 The decoder may choose to generate the floor curve at any appropriate
553 time. It is reasonable to generate the output curve when the floor
554 data is decoded from the raw packet, or it can be generated after
555 inverse coupling and applied to the spectral residue directly,
556 combining generation and the dot product into one step and eliminating
557 some working space.</para>
560 Both floor 0 and floor 1 generate a linear-range, linear-domain output
561 vector to be multiplied (dot product) by the linear-range,
562 linear-domain spectral residue.</para>
566 <section><title>compute floor/residue dot product</title>
569 This step is straightforward; for each output channel, the decoder
570 multiplies the floor curve and residue vectors element by element,
571 producing the finished audio spectrum of each channel.</para>
574 One point is worth mentioning about this dot product; a common mistake
575 in a fixed point implementation might be to assume that a 32 bit
576 fixed-point representation for floor and residue and direct
577 multiplication of the vectors is sufficient for acceptable spectral
578 depth in all cases because it happens to mostly work with the current
579 Xiph.Org reference encoder.</para>
582 However, floor vector values can span ~140dB (~24 bits unsigned), and
583 the audio spectrum vector should represent a minimum of 120dB (~21
584 bits with sign), even when output is to a 16 bit PCM device. For the
585 residue vector to represent full scale if the floor is nailed to
586 -140dB, it must be able to span 0 to +140dB. For the residue vector
587 to reach full scale if the floor is nailed at 0dB, it must be able to
588 represent -140dB to +0dB. Thus, in order to handle full range
589 dynamics, a residue vector may span -140dB to +140dB entirely within
590 spec. A 280dB range is approximately 48 bits with sign; thus the
591 residue vector must be able to represent a 48 bit range and the dot
592 product must be able to handle an effective 48 bit times 24 bit
593 multiplication. This range may be achieved using large (64 bit or
594 larger) integers, or implementing a movable binary point
595 representation.</para>
599 <section><title>inverse monolithic transform (MDCT)</title>
602 The audio spectrum is converted back into time domain PCM audio via an
603 inverse Modified Discrete Cosine Transform (MDCT). A detailed
604 description of the MDCT is available in the paper <ulink
605 url="http://www.iocon.com/resource/docs/ps/eusipco_corrected.ps"><citetitle pubwork="article">The use of multirate filter banks for coding of high quality digital
606 audio</citetitle></ulink>, by T. Sporer, K. Brandenburg and B. Edler.</para>
609 Note that the PCM produced directly from the MDCT is not yet finished
610 audio; it must be lapped with surrounding frames using an appropriate
611 window (such as the Vorbis window) before the MDCT can be considered
616 <section><title>overlap/add data</title>
618 Windowed MDCT output is overlapped and added with the right hand data
619 of the previous window such that the 3/4 point of the previous window
620 is aligned with the 1/4 point of the current window (as illustrated in
621 the window overlap diagram). At this point, the audio data between the
622 center of the previous frame and the center of the current frame is
623 now finished and ready to be returned. </para>
626 <section><title>cache right hand data</title>
628 The decoder must cache the right hand portion of the current frame to
629 be lapped with the left hand portion of the next frame.
633 <section><title>return finished audio data</title>
636 The overlapped portion produced from overlapping the previous and
637 current frame data is finished data to be returned by the decoder.
638 This data spans from the center of the previous window to the center
639 of the current window. In the case of same-sized windows, the amount
640 of data to return is one-half block consisting of and only of the
641 overlapped portions. When overlapping a short and long window, much of
642 the returned range is not actually overlap. This does not damage
643 transform orthogonality. Pay attention however to returning the
644 correct data range; the amount of data to be returned is:
647 window_blocksize(previous_window)/4+window_blocksize(current_window)/4
650 from the center of the previous window to the center of the current
654 Data is not returned from the first frame; it must be used to 'prime'
655 the decode engine. The encoder accounts for this priming when
656 calculating PCM offsets; after the first frame, the proper PCM output
657 offset is '0' (as no data has been returned yet).</para>
664 <!-- end Vorbis I specification introduction and description -->