1 <HTML><HEAD><TITLE>xiph.org: Ogg Vorbis documentation</TITLE>
2 <BODY bgcolor="#ffffff" text="#202020" link="#006666" vlink="#000000">
3 <nobr><img src="white-ogg.png"><img src="vorbisword2.png"></nobr><p>
5 <h1><font color=#000070>
6 Ogg Vorbis I format specification: probability model and codebooks
9 <em>Last update to this document: August 8, 2002</em><br>
13 Unlike practically every other mainstream audio codec, Vorbis has no
14 statically configured probability model, instead packing all entropy
15 decoding configuration, VQ and Huffman, into the bitstream itself in
16 the third header, the codec setup header. This packed configuration
17 consists of multiple 'codebooks', each containing a specific
18 Huffman-equivalent representation for decoding compressed codewords as
19 well as an optional lookup table of output vector values to which a
20 decoded Huffman value is applied as an offset, generating the final
21 decoded output corresponding to a given compressed codeword.
23 <h2>bitwise operation</h2>
25 The codebook mechanism is built on top of the <a
26 href="vorbis-spec-bitpack.html">Vorbis bitpacker</a>; both the
27 codebooks themselves and the codewords they decode are unrolled from a
28 packet as a series of arbitrary-width values read from the stream
29 according to the <a href="vorbis-spec-bitpack.html">Vorbis bitpacking
32 <h1>Packed Codebook Format</h1>
34 For purposes of the below examples, we assume that the storage
35 system's native byte width is eight bits. This is not universally
36 true; see <a href="vorbis-spec-bitpack.html">the Vorbis bitpacking
37 convention</a> document for discussion relating to non-eight-bit
40 <h2>codebook decode</h2>
42 A codebook begins with a 24 bit sync pattern, 0x564342:
45 byte 0: [ 0 1 0 0 0 0 1 0 ] (0x42)
46 byte 1: [ 0 1 0 0 0 0 1 1 ] (0x43)
47 byte 2: [ 0 1 0 1 0 1 1 0 ] (0x56)
50 16 bit <tt>[codebook_dimensions]</tt> and 24 bit <tt>[codebook_entries]</tt> fields:
54 byte 3: [ X X X X X X X X ]
55 byte 4: [ X X X X X X X X ] [codebook_dimensions] (16 bit unsigned)
57 byte 5: [ X X X X X X X X ]
58 byte 6: [ X X X X X X X X ]
59 byte 7: [ X X X X X X X X ] [codebook_entries] (24 bit unsigned)
63 Next is the <tt>[ordered]</tt> bit flag:
67 byte 8: [ X ] [ordered] (1 bit)
71 Each entry, numbering a
72 total of <tt>[codebook_entries]</tt>, is assigned a codeword length.
73 We now read the list of codeword lengths and store these lengths in
74 the array <tt>[codebook_codeword_lengths]</tt>. Decode of lengths is
75 according to whether the <tt>[ordered]</tt> flag is set or unset.
79 <li>If the <tt>[ordered]</tt> flag is unset, the codeword list is not
80 length ordered and the decoder needs to read each codeword length
81 one-by-one.<p> The decoder first reads one additional bit flag, the
82 <tt>[sparse]</tt> flag. This flag determines whether or not the
83 codebook contains unused entries that are not to be included in the
84 codeword decode tree:<p>
87 byte 8: [ X 1 ] [sparse] flag (1 bit)
90 The decoder now performs for each of the <tt>[codebook_entries]</tt> code book entries:
94 1) if([sparse] is set){
96 2) [flag] = read one bit;
99 4) [length] = read a five bit unsigned integer;
100 5) codeword length for this entry is [length]+1;
104 6) this entry is unused. mark it as such.
108 } else the sparse flag is not set {
110 7) [length] = read a five bit unsigned integer;
111 8) the codeword length for this entry is [length]+1;
117 <li>If the <tt>[ordered]</tt> flag is set, the codeword list for this
118 codebook is encoded in ascending length order. Rather than reading
119 a length for every codeword, the encoder reads the number of
120 codewords per length. That is, beginning at entry zero:
123 1) [current_entry] = 0;
124 2) [current_length] = read a five bit unsigned integer and add 1;
125 3) [number] = read <a
126 href="helper.html#ilog">ilog</a>([codebook_entries] - [current_entry]) bits as an unsigned integer
127 4) set the entries [current_entry] through [current_entry]+[number]-1, inclusive,
128 of the [codebook_codeword_lengths] array to [current_length]
129 5) set [current_entry] to [number] + [current_entry]
130 6) increment [current_length] by 1
131 7) if [current_entry] is greater than [codebook_entries] ERROR CONDITION; the decoder will
132 not be able to read this stream.
133 8) if [current_entry] is less than [codebook_entries], repeat process starting at 3)
139 After all codeword lengths have been decoded, the decoder reads the
140 vector lookup table. Vorbis I supports three lookup types:
142 <li>Implicitly populated value mapping (lattice VQ)
143 <li>Explicitly populated value mapping (tessellated or 'foam' VQ)
146 The lookup table type is read as a four bit unsigned integer:
148 1) [codebook_lookup_type] = read four bits as an unsigned integer
151 Codebook decode precedes according to <tt>[codebook_lookup_type]</tt>:
153 <li> Lookup type zero indicates no lookup to be read. Proceed past
156 <li> Lookup types one and two are similar, differing only in the
157 number of lookup values to be read. Lookup type one reads a list of
158 values that are permuted in a set pattern to build a list of vectors,
159 each vector of order <tt>[codebook_dimensions]</tt> scalars. Lookup
160 type two builds the same vector list, but reads each scalar for each
161 vector explicitly, rather than building vectors from a smaller list of
162 possible scalar values. Lookup decode proceeds as follows:
165 1) [codebook_minimum_value] = <a href="helper.html#float32_unpack">float32_unpack</a>( read 32 bits as an unsigned integer)
166 2) [codebook_delta_value] = <a href="helper.html#float32_unpack">float32_unpack</a>( read 32 bits as an unsigned integer)
167 3) [codebook_value_bits] = read 4 bits as an unsigned integer and add 1
168 4) [codebook_sequence_p] = read 1 bit as a boolean flag
170 if ( [codebook_lookup_type] is 1 ) {
172 5) [codebook_lookup_values] = <a href="helper.html#lookup1_values">lookup1_values</a>( <tt>[codebook_entries]</tt>, <tt>[codebook_dimensions]</tt> )
176 6) [codebook_lookup_values] = <tt>[codebook_entries]</tt> * <tt>[codebook_dimensions]</tt>
180 7) read a total of [codebook_lookup_values] unsigned integers of [codebook_value_bits] each;
181 store these in order in the array [codebook_multiplicands]
184 <li>A <tt>[codebook_lookup_type]</tt> of greater than two is reserved and indicates
185 a stream that's not decodable by the specification in this document.
189 An 'end of packet' during any read operation in the above steps is
190 considered an error condition rendering the stream undecodable.<p>
192 <h2>Huffman decision tree representation</h2>
194 The <tt>[codebook_codeword_lengths]</tt> array and
195 <tt>[codebook_entries]</tt> value uniquely define the Huffman decision
196 tree used for entropy decoding.<p>
198 Briefly, each used codebook entry (recall that length-unordered
199 codebooks support unused codeword entries) is assigned, in order, the
200 lowest valued unused binary Huffman codeword possible. Assume the
201 following codeword length list:<p>
214 Assigning codewords in order (lowest possible value of the appropriate
215 length to highest) results in the following codeword list:<p>
218 entry 0: length 2 codeword 00
219 entry 1: length 4 codeword 0100
220 entry 2: length 4 codeword 0101
221 entry 3: length 4 codeword 0110
222 entry 4: length 4 codeword 0111
223 entry 5: length 2 codeword 10
224 entry 6: length 3 codeword 110
225 entry 7: length 3 codeword 111
228 <em>note that unlike most binary numerical values in this document, we
229 intend the above codewords to be read and used bit by bit from left to
230 right, thus the codeword '001' is the bit string 'zero, zero, one'.
231 When determining 'lowest possible value' in the assignment definition
232 above, the leftmost bit is the MSb.</em><p>
234 It is clear that the codeword length list represents a Huffman
235 decision tree with the entry numbers equivalent to the leaves numbered
238 <img src="hufftree.png"><p>
240 As we assign codewords in order, we see that each choice constructs a
241 new leaf in the leftmost possible position.<p>
243 Note that it's possible to underspecify or overspecify a Huffman tree
244 via the length list. In the above example, if codeword seven were
245 eliminated, it's clear that the tree is unfinished:<p>
247 <img src="hufftree-under.png"><p>
249 Similarly, in the original codebook, it's clear that the tree is fully
250 populated and a ninth codeword is impossible. Both underspecified and
251 overspecified trees are an error condition rendering the stream
254 Codebook entries marked 'unused' are simply skipped in the assigning
255 process. They have no codeword and do not appear in the decision
256 tree, thus it's impossible for any bit pattern read from the stream to
257 decode to that entry number.<p>
259 <h2>VQ lookup table vector representation</h2>
261 Unpacking the VQ lookup table vectors relies on the following values:
263 <li> the <tt>[codebook_multiplicands]</tt> array
264 <li> <tt>[codebook_minimum_value]</tt>
265 <li> <tt>[codebook_delta_value]</tt>
266 <li> <tt>[codebook_sequence_p]</tt>
267 <li> <tt>[codebook_lookup_type]</tt>
268 <li> <tt>[codebook_entries]</tt>
269 <li> <tt>[codebook_dimensions]</tt>
270 <li> <tt>[codebook_lookup_values]</tt>
274 Decoding [unpacking] a specific vector in the vector lookup table
275 proceeds according to <tt>[codebook_lookup_type]</tt>. The unpacked
276 vector values are what a codebook would return during audio packet
277 decode in a VQ context.<p>
279 <h3>Vector value decode: Lookup type 1</h3>
281 Lookup type one specifies a lattice VQ lookup table built
282 algorithmically from a list of scalar values. Calculate [unpack] the
283 final values of a codebook entry vector from the entries in
284 <tt>[codebook_multiplicands]</tt> as follows (<tt>[value_vector]</tt>
285 is the output vector representing the vctor of values for entry number
286 <tt>[lookup_offset]</tt> in this codebook):<p>
290 2) [index_divisor] = 1;
291 3) iterate [i] over the range 0 ... [codebook_dimensions]-1 (once for each scalar value in the value vector) {
293 4) [multiplicand_offset] = ( [lookup_offset] divided by [index_divisor] using integer
294 division ) integer modulo [codebook_lookup_values]
296 5) vector [value_vector] element [i] =
297 ( [codebook_multiplicands] array element number [multiplicand_offset] ) *
298 [codebook_delta_value] + [codebook_minimum_value] + [last];
300 6) if ( [codebook_sequence_p] is set ) then set [last] = vector [value_vector] element [i]
302 7) [index_divisor] = [index_divisor] * [codebook_lookup_values]
306 8) vector calculation completed.
309 <h3>Vector value decode: Lookup type 2</h3>
311 Lookup type two specifies a VQ lookup table in which each scalar in
312 each vector is explicitly set by the <tt>[codebook_multiplicands]</tt>
313 array in a one-to-one mapping. Calculate [unpack] the
314 final values of a codebook entry vector from the entries in
315 <tt>[codebook_multiplicands]</tt> as follows (<tt>[value_vector]</tt>
316 is the output vector representing the vctor of values for entry number
317 <tt>[lookup_offset]</tt> in this codebook):<p>
321 2) [multiplicand_offset] = [lookup_offset] * [codebook_dimensions]
322 3) iterate [i] over the range 0 ... [codebook_dimensions]-1 (once for each scalar value in the value vector) {
324 4) vector [value_vector] element [i] =
325 ( [codebook_multiplicands] array element number [multiplicand_offset] ) *
326 [codebook_delta_value] + [codebook_minimum_value] + [last];
328 5) if ( [codebook_sequence_p] is set ) then set [last] = vector [value_vector] element [i]
330 6) increment [multiplicand_offset]
334 7) vector calculation completed.
337 <h1>Use of the codebook abstraction</h1>
339 The decoder uses the codebook abstraction much as it does the
340 bit-unpacking convention; a specific codebook reads a
341 codeword from the bitstream, decoding it into an entry number, and then
342 returns that entry number to the decoder (when used in a scalar
343 entropy coding context), or uses that entry number as an offset into
344 the VQ lookup table, returning a vector of values (when used in a context
345 desiring a VQ value). Scalar or VQ context is always explicit; any call
346 to the codebook mechanism requests either a scalar entry number or a
349 Note that VQ lookup type zero indicates that there is no lookup table;
350 requesting decode using a codebook of lookup type 0 in any context
351 expecting a vector return value (even in a case where a vector of
352 dimension one) is forbidden. If decoder setup or decode requests such
353 an action, that is an error condition rendering the packet
356 Using a codebook to read from the packet bitstream consists first of
357 reading and decoding the next codeword in the bitstream. The decoder
358 reads bits until the accumulated bits match a codeword in the
359 codebook. This process can be though of as logically walking the
360 Huffman decode tree by reading one bit at a time from the bitstream,
361 and using the bit as a decision boolean to take the 0 branch (left in
362 the above examples) or the 1 branch (right in the above examples).
363 Walking the tree finishes when the decode process hits a leaf in the
364 decision tree; the result is the entry number corresponding to that
365 leaf. Reading past the end of a packet propagates the 'end-of-stream'
366 condition to the decoder.<p>
368 When used in a scalar context, the resulting codeword entry is the
369 desired return value.<p>
371 When used in a VQ context, the codeword entry number is used as an
372 offset into the VQ lookup table. The value returned to the decoder is
373 the vector of scalars corresponding to this offset.<p>
376 <a href="http://www.xiph.org/">
377 <img src="white-xifish.png" align=left border=0>
379 <font size=-2 color=#505050>
381 Ogg is a <a href="http://www.xiph.org">Xiph.org Foundation</a> effort
382 to protect essential tenets of Internet multimedia from corporate
383 hostage-taking; Open Source is the net's greatest tool to keep
384 everyone honest. See <a href="http://www.xiph.org/about.html">About
385 the Xiph.org Foundation</a> for details.
388 Ogg Vorbis is the first Ogg audio CODEC. Anyone may freely use and
389 distribute the Ogg and Vorbis specification, whether in a private,
390 public or corporate capacity. However, the Xiph.org Foundation and
391 the Ogg project (xiph.org) reserve the right to set the Ogg Vorbis
392 specification and certify specification compliance.<p>
394 Xiph.org's Vorbis software CODEC implementation is distributed under a
395 BSD-like license. This does not restrict third parties from
396 distributing independent implementations of Vorbis software under
399 Ogg, Vorbis, Xiph.org Foundation and their logos are trademarks (tm)
400 of the <a href="http://www.xiph.org/">Xiph.org Foundation</a>. These
401 pages are copyright (C) 1994-2002 Xiph.org Foundation. All rights