4 liba52 provides a low-level interface to decoding audio frames encoded
5 using ATSC standard A/52 aka AC-3. liba52 provides downmixing and
6 dynamic range compression for the following output configurations:
8 A52_CHANNEL : Dual mono. Two independant mono channels.
9 A52_CHANNEL1 : First of the two mono channels above.
10 A52_CHANNEL2 : Second of the two mono channels above.
13 A52_DOLBY : Dolby surround compatible stereo.
14 A52_3F : 3 front channels (left, center, right)
15 A52_2F1R : 2 front, 1 rear surround channel (L, R, S)
16 A52_3F1R : 3 front, 1 rear surround channel (L, C, R, S)
17 A52_2F2R : 2 front, 2 rear surround channels (L, R, LS, RS)
18 A52_3F2R : 3 front, 2 rear surround channels (L, C, R, LS, RS)
20 A52_LFE : Low frequency effects channel. Normally used to connect a
21 subwoofer. Can be combined with any of the above channels.
22 For example: A52_3F2R | A52_LFE -> 3 front, 2 rear, 1 LFE (5.1)
28 sample_t * a52_init (uint32_t mm_accel);
30 Initializes the A/52 library. Takes as a parameter the acceptable
31 optimizations which may be used, such as MMX. These are found in the
32 included header file 'mm_accel', along with an autodetection function
33 (mm_accel()). Currently, the only accelleration implemented is
34 MM_ACCEL_MLIB, which uses the 'mlib' library if installed. mlib is
35 only available on some Sun Microsystems platforms.
37 The return value is a pointer to a properly-aligned sample buffer used
44 int a52_syncinfo (uint8_t * buf, int * flags,
45 int * sample_rate, int * bit_rate);
47 The A/52 bitstream is composed of several a52 frames concatenated one
48 after each other. An a52 frame is the smallest independantly decodable
51 buf must contain at least 7 bytes from the input stream. If these look
52 like the start of a valid a52 frame, a52_syncinfo() returns the size
53 of the coded frame in bytes, and fills flags, sample_rate and bit_rate
54 with the information encoded in the stream. The returned size is
55 guaranteed to be an even number between 128 and 3840. sample_rate will
56 be the sampling frequency in Hz, bit_rate is for the compressed stream
57 and is in bits per second, and flags is a description of the coded
58 channels: the A52_LFE bit is set if there is an LFE channel coded in
59 this stream, and by masking flags with A52_CHANNEL_MASK you will get a
60 value that describes the full-bandwidth channels, as one of the
61 A52_CHANNEL...A52_3F2R flags.
63 If this can not possibly be a valid frame, then the function returns
64 0. You should then try to re-synchronize with the a52 stream - one way
65 to try this would be to advance buf by one byte until its contents
66 looks like a valid frame, but there might be better
67 application-specific ways to synchronize.
69 It is recommended to call this function for each frame, for several
70 reasons: this function detects errors that the other functions will
71 not double-check, consecutive frames might have different lengths, and
72 it helps you re-sync with the stream if you get de-synchronized.
75 Starting to decode a frame
76 --------------------------
78 int a52_frame (a52_state_t * state, uint8_t * buf, int * flags,
79 sample_t * level, sample_t bias);
81 This starts the work of decoding the A/52 frame (to be completed using
82 a52_block()). buf should point to the beginning of the complete frame
83 of the full size returned by a52_syncinfo().
85 You should pass in the flags the speaker configuration that you
86 support, and liba52 will return the speaker configuration it will use
87 for its output, based on what is coded in the stream and what you
88 asked for. For example, if the stream contains 2+2 channels
89 (a52_syncinfo() returned A52_2F2R in the flags), and you have 3+1
90 speakers (you passed A52_3F1R), then liba52 will choose do downmix to
91 2+1 speakers, since there is no center channel to send to your center
92 speaker. So in that case the left and right channels will be
93 essentially unmodified by the downmix, and the two surround channels
94 will be added together and sent to your surround speaker. liba52 will
95 return A52_2F1R to indicate this.
97 The good news is that when you downmix to stereo you dont have to
98 worry about this, you will ALWAYS get a stereo output no matter what
99 was coded in the stream. For more complex output configurations you
100 will have to handle the case where liba52 couldnt give you what you
101 wanted because some of the channels were not encoded in the stream
104 Level, bias, and A52_ADJUST_LEVEL:
106 Before downmixing, samples are floating point values with a range of
107 [-1,1]. Most types of downmixing will combine channels together, which
108 will potentially result in a larger range for the output
109 samples. liba52 provides two methods of controlling the range of the
110 output, either before or after the downmix stage.
112 If you do not set A52_ADJUST_LEVEL, liba52 will multiply the samples
113 by your level value, so that they fit in the [-level,level]
114 range. Then it will apply the standardized downmix equations,
115 potentially making the samples go out of that interval again. The
116 level parameter is not modified.
118 Setting the A52_ADJUST_LEVEL flag will instruct liba52 to treat your
119 level value as the intended range interval after downmixing. It will
120 then figure out what level to use before the downmix (what you should
121 have passed if you hadnt used the A52_ADJUST_LEVEL flag), and
122 overwrite the level value you gave it with that new level value.
124 The bias represents a value which should be added to the result
127 output_sample = (input_sample * level) + bias;
129 For example, a bias of 384 and a level of 1 tells liba52 you want
130 samples between 383 and 385 instead of -1 and 1. This is what the
131 sample program a52dec does, as it makes it faster to convert the
132 samples to integer format, using a trick based on the IEEE
133 floating-point format.
135 This function also initialises the state for that frame, which will be
136 reused next when decoding blocks.
139 Dynamic range compression
140 -------------------------
142 void a52_dynrng (a52_state_t * state,
143 sample_t (* call) (sample_t, void *), void * data);
145 This function is purely optional. If you dont call it, liba52 will
146 provide the default behaviour, which is to apply the full dynamic
147 range compression as specified in the A/52 stream. This basically
148 makes the loud sounds softer, and the soft sounds louder, so you can
149 more easily listen to the stream in a noisy environment without
152 If you do call this function and set a NULL callback, this will
153 totally disable the dynamic range compression and provide a playback
154 more adapted to a movie theater or a listening room.
156 If you call this function and specify a callback function, this
157 callback might be called up to once for each block, with two
158 arguments: the compression factor 'c' recommended by the bitstream,
159 and the private data pointer you specified in a52_dynrng(). The
160 callback will then return the amount of compression to actually use -
161 typically pow(c,x) where x is somewhere between 0 and 1. More
162 elaborate compression functions might want to use a different value
163 for 'x' depending wether c>1 or c<1 - or even something more complex
164 if this is what you want.
170 int a52_block (a52_state_t * state, sample_t * samples);
172 Every A/52 frame is composed of 6 blocks, each with an output of 256
173 samples for each channel. The a52_block() function decodes the next
174 block in the frame, and should be called 6 times to decode all of the
175 audio in the frame. After each call, you should extract the audio data
176 from the sample buffer.
178 The sample pointer given should be the one a52_init() returned.
180 After this function returns, the samples buuffer will contain 256
181 samples for the first channel, followed by 256 samples for the second
182 channel, etc... the channel order is LFE, left, center, right, left
183 surround, right surround. If one of the channels is not present in the
184 liba52 output, as indicated by the flags returned by a52_frame(), then
185 this channel is skipped and the following channels are shifted so
186 liba52 does not leave an empty space between channels.
192 sample_t * samples = a52_init (mm_accel());
195 if at least 7 bytes in the buffer:
197 bytes_to_get = a52_syncinfo (...)
199 if bytes_to_get == 0:
200 goto loop to keep looking for sync point
204 a52_frame (state, buf, ...)
205 [a52_dynrng (state, ...); this is only optional]
207 a52_block (state, samples)
208 convert samples to integer and queue to soundcard