many doc cleanups
[platform/upstream/libvorbis.git] / doc / vorbis-fidelity.html
1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">\r
2 <html>\r
3 <head>\r
4 \r
5 <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-15"/>\r
6 <title>Ogg Vorbis Documentation</title>\r
7 \r
8 <style type="text/css">\r
9 body {\r
10   margin: 0 18px 0 18px;\r
11   padding-bottom: 30px;\r
12   font-family: Verdana, Arial, Helvetica, sans-serif;\r
13   color: #333333;\r
14   font-size: .8em;\r
15 }\r
16 \r
17 a {\r
18   color: #3366cc;\r
19 }\r
20 \r
21 img {\r
22   border: 0;\r
23 }\r
24 \r
25 #xiphlogo {\r
26   margin: 30px 0 16px 0;\r
27 }\r
28 \r
29 #content p {\r
30   line-height: 1.4;\r
31 }\r
32 \r
33 h1, h1 a, h2, h2 a, h3, h3 a {\r
34   font-weight: bold;\r
35   color: #ff9900;\r
36   margin: 1.3em 0 8px 0;\r
37 }\r
38 \r
39 h1 {\r
40   font-size: 1.3em;\r
41 }\r
42 \r
43 h2 {\r
44   font-size: 1.2em;\r
45 }\r
46 \r
47 h3 {\r
48   font-size: 1.1em;\r
49 }\r
50 \r
51 li {\r
52   line-height: 1.4;\r
53 }\r
54 \r
55 #copyright {\r
56   margin-top: 30px;\r
57   line-height: 1.5em;\r
58   text-align: center;\r
59   font-size: .8em;\r
60   color: #888888;\r
61   clear: both;\r
62 }\r
63 </style>\r
64 \r
65 </head>\r
66 \r
67 <body>\r
68 \r
69 <div id="xiphlogo">\r
70   <a href="http://www.xiph.org/"><img src="fish_xiph_org.png" alt="Fish Logo and Xiph.org"/></a>\r
71 </div>\r
72 \r
73 <h1>Ogg Vorbis: Fidelity measurement and terminology discussion</h1>\r
74 \r
75 <p>Terminology discussed in this document is based on common terminology\r
76 associated with contemporary codecs such as MPEG I audio layer 3\r
77 (mp3). However, some differences in terminology are useful in the\r
78 context of Vorbis as Vorbis functions somewhat differently than most\r
79 current formats. For clarity, then, we describe a common terminology\r
80 for discussion of Vorbis's and other formats' audio quality.</p>\r
81 \r
82 <h2>Subjective and Objective</h2>\r
83 \r
84 <p><em>Objective</em> fidelity is a measure, based on a computable,\r
85 mechanical metric, of how carefully an output matches an input. For\r
86 example, a stereo amplifier may claim to introduce less that .01%\r
87 total harmonic distortion when amplifying an input signal; this claim\r
88 is easy to verify given proper equipment, and any number of testers are\r
89 likely to arrive at the same, exact results. One need not listen to\r
90 the equipment to make this measurement.</p>\r
91 \r
92 <p>However, given two amplifiers with identical, verifiable objective\r
93 specifications, listeners may strongly prefer the sound quality of one\r
94 over the other. This is actually the case in the decades old debate\r
95 [some would say jihad] among audiophiles involving vacuum tube versus\r
96 solid state amplifiers. There are people who can tell the difference,\r
97 and strongly prefer one over the other despite seemingly identical,\r
98 measurable quality. This preference is <em>subjective</em> and\r
99 difficult to measure but nonetheless real.</p>\r
100 \r
101 <p>Individual elements of subjective differences often can be qualified,\r
102 but overall subjective quality generally is not measurable. Different\r
103 observers are likely to disagree on the exact results of a subjective\r
104 test as each observer's perspective differs. When measuring\r
105 subjective qualities, the best one can hope for is average, empirical\r
106 results that show statistical significance across a group.</p>\r
107 \r
108 <p>Perceptual codecs are most concerned with subjective, not objective,\r
109 quality. This is why evaluating a perceptual codec via distortion\r
110 measures and sonograms alone is useless; these objective measures may\r
111 provide insight into the quality or functioning of a codec, but cannot\r
112 answer the much squishier subjective question, "Does it sound\r
113 good?". The tube amplifier example is perhaps not the best as very few\r
114 people can hear, or care to hear, the minute differences between tubes\r
115 and transistors, whereas the subjective differences in perceptual\r
116 codecs tend to be quite large even when objective differences are\r
117 not.</p>\r
118 \r
119 <h2>Fidelity, Artifacts and Differences</h2>\r
120 \r
121 <p>Audio <em>artifacts</em> and loss of fidelity or more simply\r
122 put, audio <em>differences</em> are not the same thing.</p>\r
123 \r
124 <p>A loss of fidelity implies differences between the perceived input and\r
125 output signal; it does not necessarily imply that the differences in\r
126 output are displeasing or that the output sounds poor (although this\r
127 is often the case). Tube amplifiers are <em>not</em> higher fidelity\r
128 than modern solid state and digital systems. They simply produce a\r
129 form of distortion and coloring that is either unnoticeable or actually\r
130 pleasing to many ears.</p>\r
131 \r
132 <p>As compared to an original signal using hard metrics, all perceptual\r
133 codecs [ASPEC, ATRAC, MP3, WMA, AAC, TwinVQ, AC3 and Vorbis included]\r
134 lose objective fidelity in order to reduce bitrate. This is fact. The\r
135 idea is to lose fidelity in ways that cannot be perceived. However,\r
136 most current streaming applications demand bitrates lower than what\r
137 can be achieved by sacrificing only objective fidelity; this is also\r
138 fact, despite whatever various company press releases might claim.\r
139 Subjective fidelity eventually must suffer in one way or another.</p>\r
140 \r
141 <p>The goal is to choose the best possible tradeoff such that the\r
142 fidelity loss is graceful and not obviously noticeable. Most listeners\r
143 of FM radio do not realize how much lower fidelity that medium is as\r
144 compared to compact discs or DAT. However, when compared directly to\r
145 source material, the difference is obvious. A cassette tape is lower\r
146 fidelity still, and yet the degradation, relatively speaking, is\r
147 graceful and generally easy not to notice. Compare this graceful loss\r
148 of quality to an average 44.1kHz stereo mp3 encoded at 80 or 96kbps.\r
149 The mp3 might actually be higher objective fidelity but subjectively\r
150 sounds much worse.</p>\r
151 \r
152 <p>Thus, when a CODEC <em>must</em> sacrifice subjective quality in order\r
153 to satisfy a user's requirements, the result should be a\r
154 <em>difference</em> that is generally either difficult to notice\r
155 without comparison, or easy to ignore. An <em>artifact</em>, on the\r
156 other hand, is an element introduced into the output that is\r
157 immediately noticeable, obviously foreign, and undesired. The famous\r
158 'underwater' or 'twinkling' effect synonymous with low bitrate (or\r
159 poorly encoded) mp3 is an example of an <em>artifact</em>. This\r
160 working definition differs slightly from common usage, but the coined\r
161 distinction between differences and artifacts is useful for our\r
162 discussion.</p>\r
163 \r
164 <p>The goal, when it is absolutely necessary to sacrifice subjective\r
165 fidelity, is obviously to strive for differences and not artifacts.\r
166 The vast majority of codecs today fail at this task miserably,\r
167 predictably, and regularly in one way or another. Avoiding such\r
168 failures when it is necessary to sacrifice subjective quality is a\r
169 fundamental design objective of Vorbis and that objective is reflected\r
170 in Vorbis's design and tuning.</p>\r
171 \r
172 <div id="copyright">\r
173   The Xiph Fish Logo is a\r
174   trademark (&trade;) of Xiph.Org.<br/>\r
175 \r
176   These pages &copy; 1994 - 2005 Xiph.Org. All rights reserved.\r
177 </div>\r
178 \r
179 </body>\r
180 </html>\r