xref: /aosp_15_r20/external/libopus/README (revision a58d3d2adb790c104798cd88c8a3aff4fa8b82cc)
1== Opus audio codec ==
2
3Opus is a codec for interactive speech and audio transmission over the Internet.
4
5  Opus can handle a wide range of interactive audio applications, including
6Voice over IP, videoconferencing, in-game  chat, and even remote live music
7performances. It can scale from low bit-rate narrowband speech to very high
8quality stereo music.
9
10  Opus, when coupled with an appropriate container format, is also suitable
11for non-realtime  stored-file applications such as music distribution, game
12soundtracks, portable music players, jukeboxes, and other applications that
13have historically used high latency formats such as MP3, AAC, or Vorbis.
14
15                    Opus is specified by IETF RFC 6716:
16                    https://tools.ietf.org/html/rfc6716
17
18  The Opus format and this implementation of it are subject to the royalty-
19free patent and copyright licenses specified in the file COPYING.
20
21This package implements a shared library for encoding and decoding raw Opus
22bitstreams. Raw Opus bitstreams should be used over RTP according to
23 https://tools.ietf.org/html/rfc7587
24
25The package also includes a number of test tools used for testing the
26correct operation of the library. The bitstreams read/written by these
27tools should not be used for Opus file distribution: They include
28additional debugging data and cannot support seeking.
29
30Opus stored in files should use the Ogg encapsulation for Opus which is
31described at:
32 https://tools.ietf.org/html/rfc7845
33
34An opus-tools package is available which provides encoding and decoding of
35Ogg encapsulated Opus files and includes a number of useful features.
36
37Opus-tools can be found at:
38 https://gitlab.xiph.org/xiph/opus-tools.git
39or on the main Opus website:
40 https://opus-codec.org/
41
42== Deep Learning and Opus ==
43
44Lossy networks continue to be a challenge for real-time communications.
45While the original implementation of Opus provides an excellent packet loss
46concealment mechanism, the team has continued to advance the methodology used
47to improve audio quality in challenge network environments.
48
49In Opus 1.5, we added a deep learning based redundancy encoder that enhances
50audio in lossy networks by embedding one second of recovery data in the padding
51data of each packet. The underlying algorithm behind encoding and decoding the
52recovery data is called the deep redundancy (DRED) algorithm. By leveraging
53the padding data within the packet, Opus 1.5 is fully backward compatible with
54prior revisions of Opus. Please see the README under the "dnn" subdirectory to
55understand DRED.
56
57DRED was developed by a team that Amazon Web Services initially sponsored,
58who open-sourced the implementation as well as began the
59standardization process at the IETF:
60  https://datatracker.ietf.org/doc/draft-ietf-mlcodec-opus-extension/
61The license behind Opus or the intellectual property position of Opus does
62not change with Opus 1.5.
63
64== Compiling libopus ==
65
66To build from a distribution tarball, you only need to do the following:
67
68    % ./configure
69    % make
70
71To build from the git repository, the following steps are necessary:
72
730) Set up a development environment:
74
75On an Ubuntu or Debian family Linux distribution:
76
77    % sudo apt-get install git autoconf automake libtool gcc make
78
79On a Fedora/Redhat based Linux:
80
81    % sudo dnf install git autoconf automake libtool gcc make
82
83Or for older Redhat/Centos Linux releases:
84
85    % sudo yum install git autoconf automake libtool gcc make
86
87On Apple macOS, install Xcode and brew.sh, then in the Terminal enter:
88
89    % brew install autoconf automake libtool
90
911) Clone the repository:
92
93    % git clone https://gitlab.xiph.org/xiph/opus.git
94    % cd opus
95
962) Compiling the source
97
98    % ./autogen.sh
99    % ./configure
100    % make
101
102On x86, it's a good idea to use a -march= option that allows the use of AVX2.
103
1043) Install the codec libraries (optional)
105
106    % sudo make install
107
108Once you have compiled the codec, there will be a opus_demo executable
109in the top directory.
110
111Usage: opus_demo [-e] <application> <sampling rate (Hz)> <channels (1/2)>
112         <bits per second> [options] <input> <output>
113       opus_demo -d <sampling rate (Hz)> <channels (1/2)> [options]
114         <input> <output>
115
116mode: voip | audio | restricted-lowdelay
117options:
118  -e                : only runs the encoder (output the bit-stream)
119  -d                : only runs the decoder (reads the bit-stream as input)
120  -cbr              : enable constant bitrate; default: variable bitrate
121  -cvbr             : enable constrained variable bitrate; default:
122                      unconstrained
123  -bandwidth <NB|MB|WB|SWB|FB>
124                    : audio bandwidth (from narrowband to fullband);
125                      default: sampling rate
126  -framesize <2.5|5|10|20|40|60>
127                    : frame size in ms; default: 20
128  -max_payload <bytes>
129                    : maximum payload size in bytes, default: 1024
130  -complexity <comp>
131                    : complexity, 0 (lowest) ... 10 (highest); default: 10
132  -inbandfec        : enable SILK inband FEC
133  -forcemono        : force mono encoding, even for stereo input
134  -dtx              : enable SILK DTX
135  -loss <perc>      : simulate packet loss, in percent (0-100); default: 0
136
137input and output are little-endian signed 16-bit PCM files or opus
138bitstreams with simple opus_demo proprietary framing.
139
140== Testing ==
141
142This package includes a collection of automated unit and system tests
143which SHOULD be run after compiling the package especially the first
144time it is run on a new platform.
145
146To run the integrated tests:
147
148    % make check
149
150There is also collection of standard test vectors which are not
151included in this package for size reasons but can be obtained from:
152https://opus-codec.org/docs/opus_testvectors-rfc8251.tar.gz
153
154To run compare the code to these test vectors:
155
156    % curl -OL https://opus-codec.org/docs/opus_testvectors-rfc8251.tar.gz
157    % tar -zxf opus_testvectors-rfc8251.tar.gz
158    % ./tests/run_vectors.sh ./ opus_newvectors 48000
159
160== Compiling libopus for Windows and alternative build systems ==
161
162See cmake/README.md or meson/README.md.
163
164== Portability notes ==
165
166This implementation uses floating-point by default but can be compiled to
167use only fixed-point arithmetic by setting --enable-fixed-point (if using
168autoconf) or by defining the FIXED_POINT macro (if building manually).
169The fixed point implementation has somewhat lower audio quality and is
170slower on platforms with fast FPUs, it is normally only used in embedded
171environments.
172
173The implementation can be compiled with either a C89 or a C99 compiler.
174While it does not rely on any _undefined behavior_ as defined by C89 or
175C99, it relies on common _implementation-defined behavior_ for two's
176complement architectures:
177
178o Right shifts of negative values are consistent with two's
179  complement arithmetic, so that a>>b is equivalent to
180  floor(a/(2^b)),
181
182o For conversion to a signed integer of N bits, the value is reduced
183  modulo 2^N to be within range of the type,
184
185o The result of integer division of a negative value is truncated
186  towards zero, and
187
188o The compiler provides a 64-bit integer type (a C99 requirement
189  which is supported by most C89 compilers).
190