blob: ca7f856eacd113d8422f23e5e3c0a7d7ef5cb04b
1 | /* |
2 | * G.726 ADPCM audio codec |
3 | * Copyright (c) 2004 Roman Shaposhnik |
4 | * |
5 | * This is a very straightforward rendition of the G.726 |
6 | * Section 4 "Computational Details". |
7 | * |
8 | * This file is part of FFmpeg. |
9 | * |
10 | * FFmpeg is free software; you can redistribute it and/or |
11 | * modify it under the terms of the GNU Lesser General Public |
12 | * License as published by the Free Software Foundation; either |
13 | * version 2.1 of the License, or (at your option) any later version. |
14 | * |
15 | * FFmpeg is distributed in the hope that it will be useful, |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 | * Lesser General Public License for more details. |
19 | * |
20 | * You should have received a copy of the GNU Lesser General Public |
21 | * License along with FFmpeg; if not, write to the Free Software |
22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
23 | */ |
24 | #include <limits.h> |
25 | |
26 | #include "libavutil/channel_layout.h" |
27 | #include "libavutil/opt.h" |
28 | #include "avcodec.h" |
29 | #include "internal.h" |
30 | #include "get_bits.h" |
31 | #include "put_bits.h" |
32 | |
33 | /** |
34 | * G.726 11-bit float. |
35 | * G.726 Standard uses rather odd 11-bit floating point arithmetic for |
36 | * numerous occasions. It's a mystery to me why they did it this way |
37 | * instead of simply using 32-bit integer arithmetic. |
38 | */ |
39 | typedef struct Float11 { |
40 | uint8_t sign; /**< 1 bit sign */ |
41 | uint8_t exp; /**< 4 bits exponent */ |
42 | uint8_t mant; /**< 6 bits mantissa */ |
43 | } Float11; |
44 | |
45 | static inline Float11* i2f(int i, Float11* f) |
46 | { |
47 | f->sign = (i < 0); |
48 | if (f->sign) |
49 | i = -i; |
50 | f->exp = av_log2_16bit(i) + !!i; |
51 | f->mant = i? (i<<6) >> f->exp : 1<<5; |
52 | return f; |
53 | } |
54 | |
55 | static inline int16_t mult(Float11* f1, Float11* f2) |
56 | { |
57 | int res, exp; |
58 | |
59 | exp = f1->exp + f2->exp; |
60 | res = (((f1->mant * f2->mant) + 0x30) >> 4); |
61 | res = exp > 19 ? res << (exp - 19) : res >> (19 - exp); |
62 | return (f1->sign ^ f2->sign) ? -res : res; |
63 | } |
64 | |
65 | static inline int sgn(int value) |
66 | { |
67 | return (value < 0) ? -1 : 1; |
68 | } |
69 | |
70 | typedef struct G726Tables { |
71 | const int* quant; /**< quantization table */ |
72 | const int16_t* iquant; /**< inverse quantization table */ |
73 | const int16_t* W; /**< special table #1 ;-) */ |
74 | const uint8_t* F; /**< special table #2 */ |
75 | } G726Tables; |
76 | |
77 | typedef struct G726Context { |
78 | AVClass *class; |
79 | G726Tables tbls; /**< static tables needed for computation */ |
80 | |
81 | Float11 sr[2]; /**< prev. reconstructed samples */ |
82 | Float11 dq[6]; /**< prev. difference */ |
83 | int a[2]; /**< second order predictor coeffs */ |
84 | int b[6]; /**< sixth order predictor coeffs */ |
85 | int pk[2]; /**< signs of prev. 2 sez + dq */ |
86 | |
87 | int ap; /**< scale factor control */ |
88 | int yu; /**< fast scale factor */ |
89 | int yl; /**< slow scale factor */ |
90 | int dms; /**< short average magnitude of F[i] */ |
91 | int dml; /**< long average magnitude of F[i] */ |
92 | int td; /**< tone detect */ |
93 | |
94 | int se; /**< estimated signal for the next iteration */ |
95 | int sez; /**< estimated second order prediction */ |
96 | int y; /**< quantizer scaling factor for the next iteration */ |
97 | int code_size; |
98 | int little_endian; /**< little-endian bitstream as used in aiff and Sun AU */ |
99 | } G726Context; |
100 | |
101 | static const int quant_tbl16[] = /**< 16kbit/s 2 bits per sample */ |
102 | { 260, INT_MAX }; |
103 | static const int16_t iquant_tbl16[] = |
104 | { 116, 365, 365, 116 }; |
105 | static const int16_t W_tbl16[] = |
106 | { -22, 439, 439, -22 }; |
107 | static const uint8_t F_tbl16[] = |
108 | { 0, 7, 7, 0 }; |
109 | |
110 | static const int quant_tbl24[] = /**< 24kbit/s 3 bits per sample */ |
111 | { 7, 217, 330, INT_MAX }; |
112 | static const int16_t iquant_tbl24[] = |
113 | { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN }; |
114 | static const int16_t W_tbl24[] = |
115 | { -4, 30, 137, 582, 582, 137, 30, -4 }; |
116 | static const uint8_t F_tbl24[] = |
117 | { 0, 1, 2, 7, 7, 2, 1, 0 }; |
118 | |
119 | static const int quant_tbl32[] = /**< 32kbit/s 4 bits per sample */ |
120 | { -125, 79, 177, 245, 299, 348, 399, INT_MAX }; |
121 | static const int16_t iquant_tbl32[] = |
122 | { INT16_MIN, 4, 135, 213, 273, 323, 373, 425, |
123 | 425, 373, 323, 273, 213, 135, 4, INT16_MIN }; |
124 | static const int16_t W_tbl32[] = |
125 | { -12, 18, 41, 64, 112, 198, 355, 1122, |
126 | 1122, 355, 198, 112, 64, 41, 18, -12}; |
127 | static const uint8_t F_tbl32[] = |
128 | { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 }; |
129 | |
130 | static const int quant_tbl40[] = /**< 40kbit/s 5 bits per sample */ |
131 | { -122, -16, 67, 138, 197, 249, 297, 338, |
132 | 377, 412, 444, 474, 501, 527, 552, INT_MAX }; |
133 | static const int16_t iquant_tbl40[] = |
134 | { INT16_MIN, -66, 28, 104, 169, 224, 274, 318, |
135 | 358, 395, 429, 459, 488, 514, 539, 566, |
136 | 566, 539, 514, 488, 459, 429, 395, 358, |
137 | 318, 274, 224, 169, 104, 28, -66, INT16_MIN }; |
138 | static const int16_t W_tbl40[] = |
139 | { 14, 14, 24, 39, 40, 41, 58, 100, |
140 | 141, 179, 219, 280, 358, 440, 529, 696, |
141 | 696, 529, 440, 358, 280, 219, 179, 141, |
142 | 100, 58, 41, 40, 39, 24, 14, 14 }; |
143 | static const uint8_t F_tbl40[] = |
144 | { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6, |
145 | 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 }; |
146 | |
147 | static const G726Tables G726Tables_pool[] = |
148 | {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 }, |
149 | { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 }, |
150 | { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 }, |
151 | { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }}; |
152 | |
153 | |
154 | /** |
155 | * Paragraph 4.2.2 page 18: Adaptive quantizer. |
156 | */ |
157 | static inline uint8_t quant(G726Context* c, int d) |
158 | { |
159 | int sign, exp, i, dln; |
160 | |
161 | sign = i = 0; |
162 | if (d < 0) { |
163 | sign = 1; |
164 | d = -d; |
165 | } |
166 | exp = av_log2_16bit(d); |
167 | dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2); |
168 | |
169 | while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln) |
170 | ++i; |
171 | |
172 | if (sign) |
173 | i = ~i; |
174 | if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */ |
175 | i = 0xff; |
176 | |
177 | return i; |
178 | } |
179 | |
180 | /** |
181 | * Paragraph 4.2.3 page 22: Inverse adaptive quantizer. |
182 | */ |
183 | static inline int16_t inverse_quant(G726Context* c, int i) |
184 | { |
185 | int dql, dex, dqt; |
186 | |
187 | dql = c->tbls.iquant[i] + (c->y >> 2); |
188 | dex = (dql>>7) & 0xf; /* 4-bit exponent */ |
189 | dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */ |
190 | return (dql < 0) ? 0 : ((dqt<<dex) >> 7); |
191 | } |
192 | |
193 | static int16_t g726_decode(G726Context* c, int I) |
194 | { |
195 | int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0; |
196 | Float11 f; |
197 | int I_sig= I >> (c->code_size - 1); |
198 | |
199 | dq = inverse_quant(c, I); |
200 | |
201 | /* Transition detect */ |
202 | ylint = (c->yl >> 15); |
203 | ylfrac = (c->yl >> 10) & 0x1f; |
204 | thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint; |
205 | tr= (c->td == 1 && dq > ((3*thr2)>>2)); |
206 | |
207 | if (I_sig) /* get the sign */ |
208 | dq = -dq; |
209 | re_signal = (int16_t)(c->se + dq); |
210 | |
211 | /* Update second order predictor coefficient A2 and A1 */ |
212 | pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0; |
213 | dq0 = dq ? sgn(dq) : 0; |
214 | if (tr) { |
215 | c->a[0] = 0; |
216 | c->a[1] = 0; |
217 | for (i=0; i<6; i++) |
218 | c->b[i] = 0; |
219 | } else { |
220 | /* This is a bit crazy, but it really is +255 not +256 */ |
221 | fa1 = av_clip_intp2((-c->a[0]*c->pk[0]*pk0)>>5, 8); |
222 | |
223 | c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7); |
224 | c->a[1] = av_clip(c->a[1], -12288, 12288); |
225 | c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8); |
226 | c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]); |
227 | |
228 | for (i=0; i<6; i++) |
229 | c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8); |
230 | } |
231 | |
232 | /* Update Dq and Sr and Pk */ |
233 | c->pk[1] = c->pk[0]; |
234 | c->pk[0] = pk0 ? pk0 : 1; |
235 | c->sr[1] = c->sr[0]; |
236 | i2f(re_signal, &c->sr[0]); |
237 | for (i=5; i>0; i--) |
238 | c->dq[i] = c->dq[i-1]; |
239 | i2f(dq, &c->dq[0]); |
240 | c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */ |
241 | |
242 | c->td = c->a[1] < -11776; |
243 | |
244 | /* Update Ap */ |
245 | c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5); |
246 | c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7); |
247 | if (tr) |
248 | c->ap = 256; |
249 | else { |
250 | c->ap += (-c->ap) >> 4; |
251 | if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3)) |
252 | c->ap += 0x20; |
253 | } |
254 | |
255 | /* Update Yu and Yl */ |
256 | c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120); |
257 | c->yl += c->yu + ((-c->yl)>>6); |
258 | |
259 | /* Next iteration for Y */ |
260 | al = (c->ap >= 256) ? 1<<6 : c->ap >> 2; |
261 | c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6; |
262 | |
263 | /* Next iteration for SE and SEZ */ |
264 | c->se = 0; |
265 | for (i=0; i<6; i++) |
266 | c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]); |
267 | c->sez = c->se >> 1; |
268 | for (i=0; i<2; i++) |
269 | c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]); |
270 | c->se >>= 1; |
271 | |
272 | return av_clip(re_signal << 2, -0xffff, 0xffff); |
273 | } |
274 | |
275 | static av_cold int g726_reset(G726Context *c) |
276 | { |
277 | int i; |
278 | |
279 | c->tbls = G726Tables_pool[c->code_size - 2]; |
280 | for (i=0; i<2; i++) { |
281 | c->sr[i].mant = 1<<5; |
282 | c->pk[i] = 1; |
283 | } |
284 | for (i=0; i<6; i++) { |
285 | c->dq[i].mant = 1<<5; |
286 | } |
287 | c->yu = 544; |
288 | c->yl = 34816; |
289 | |
290 | c->y = 544; |
291 | |
292 | return 0; |
293 | } |
294 | |
295 | #if CONFIG_ADPCM_G726_ENCODER |
296 | static int16_t g726_encode(G726Context* c, int16_t sig) |
297 | { |
298 | uint8_t i; |
299 | |
300 | i = av_mod_uintp2(quant(c, sig/4 - c->se), c->code_size); |
301 | g726_decode(c, i); |
302 | return i; |
303 | } |
304 | |
305 | /* Interfacing to the libavcodec */ |
306 | |
307 | static av_cold int g726_encode_init(AVCodecContext *avctx) |
308 | { |
309 | G726Context* c = avctx->priv_data; |
310 | |
311 | if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL && |
312 | avctx->sample_rate != 8000) { |
313 | av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not " |
314 | "allowed when the compliance level is higher than unofficial. " |
315 | "Resample or reduce the compliance level.\n"); |
316 | return AVERROR(EINVAL); |
317 | } |
318 | if (avctx->sample_rate <= 0) { |
319 | av_log(avctx, AV_LOG_ERROR, "Invalid sample rate %d\n", |
320 | avctx->sample_rate); |
321 | return AVERROR(EINVAL); |
322 | } |
323 | |
324 | if(avctx->channels != 1){ |
325 | av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n"); |
326 | return AVERROR(EINVAL); |
327 | } |
328 | |
329 | if (avctx->bit_rate) |
330 | c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate; |
331 | |
332 | c->code_size = av_clip(c->code_size, 2, 5); |
333 | avctx->bit_rate = c->code_size * avctx->sample_rate; |
334 | avctx->bits_per_coded_sample = c->code_size; |
335 | |
336 | g726_reset(c); |
337 | |
338 | /* select a frame size that will end on a byte boundary and have a size of |
339 | approximately 1024 bytes */ |
340 | avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2]; |
341 | |
342 | return 0; |
343 | } |
344 | |
345 | static int g726_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
346 | const AVFrame *frame, int *got_packet_ptr) |
347 | { |
348 | G726Context *c = avctx->priv_data; |
349 | const int16_t *samples = (const int16_t *)frame->data[0]; |
350 | PutBitContext pb; |
351 | int i, ret, out_size; |
352 | |
353 | out_size = (frame->nb_samples * c->code_size + 7) / 8; |
354 | if ((ret = ff_alloc_packet2(avctx, avpkt, out_size, 0)) < 0) |
355 | return ret; |
356 | init_put_bits(&pb, avpkt->data, avpkt->size); |
357 | |
358 | for (i = 0; i < frame->nb_samples; i++) |
359 | put_bits(&pb, c->code_size, g726_encode(c, *samples++)); |
360 | |
361 | flush_put_bits(&pb); |
362 | |
363 | avpkt->size = out_size; |
364 | *got_packet_ptr = 1; |
365 | return 0; |
366 | } |
367 | |
368 | #define OFFSET(x) offsetof(G726Context, x) |
369 | #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
370 | static const AVOption options[] = { |
371 | { "code_size", "Bits per code", OFFSET(code_size), AV_OPT_TYPE_INT, { .i64 = 4 }, 2, 5, AE }, |
372 | { NULL }, |
373 | }; |
374 | |
375 | static const AVClass g726_class = { |
376 | .class_name = "g726", |
377 | .item_name = av_default_item_name, |
378 | .option = options, |
379 | .version = LIBAVUTIL_VERSION_INT, |
380 | }; |
381 | |
382 | static const AVCodecDefault defaults[] = { |
383 | { "b", "0" }, |
384 | { NULL }, |
385 | }; |
386 | |
387 | AVCodec ff_adpcm_g726_encoder = { |
388 | .name = "g726", |
389 | .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), |
390 | .type = AVMEDIA_TYPE_AUDIO, |
391 | .id = AV_CODEC_ID_ADPCM_G726, |
392 | .priv_data_size = sizeof(G726Context), |
393 | .init = g726_encode_init, |
394 | .encode2 = g726_encode_frame, |
395 | .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME, |
396 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, |
397 | AV_SAMPLE_FMT_NONE }, |
398 | .priv_class = &g726_class, |
399 | .defaults = defaults, |
400 | }; |
401 | #endif |
402 | |
403 | #if CONFIG_ADPCM_G726_DECODER || CONFIG_ADPCM_G726LE_DECODER |
404 | static av_cold int g726_decode_init(AVCodecContext *avctx) |
405 | { |
406 | G726Context* c = avctx->priv_data; |
407 | |
408 | if(avctx->channels > 1){ |
409 | avpriv_request_sample(avctx, "Decoding more than one channel"); |
410 | return AVERROR_PATCHWELCOME; |
411 | } |
412 | avctx->channels = 1; |
413 | avctx->channel_layout = AV_CH_LAYOUT_MONO; |
414 | |
415 | c->little_endian = !strcmp(avctx->codec->name, "g726le"); |
416 | |
417 | c->code_size = avctx->bits_per_coded_sample; |
418 | if (c->code_size < 2 || c->code_size > 5) { |
419 | av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size); |
420 | return AVERROR(EINVAL); |
421 | } |
422 | g726_reset(c); |
423 | |
424 | avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
425 | |
426 | return 0; |
427 | } |
428 | |
429 | static int g726_decode_frame(AVCodecContext *avctx, void *data, |
430 | int *got_frame_ptr, AVPacket *avpkt) |
431 | { |
432 | AVFrame *frame = data; |
433 | const uint8_t *buf = avpkt->data; |
434 | int buf_size = avpkt->size; |
435 | G726Context *c = avctx->priv_data; |
436 | int16_t *samples; |
437 | GetBitContext gb; |
438 | int out_samples, ret; |
439 | |
440 | out_samples = buf_size * 8 / c->code_size; |
441 | |
442 | /* get output buffer */ |
443 | frame->nb_samples = out_samples; |
444 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
445 | return ret; |
446 | samples = (int16_t *)frame->data[0]; |
447 | |
448 | init_get_bits(&gb, buf, buf_size * 8); |
449 | |
450 | while (out_samples--) |
451 | *samples++ = g726_decode(c, c->little_endian ? |
452 | get_bits_le(&gb, c->code_size) : |
453 | get_bits(&gb, c->code_size)); |
454 | |
455 | if (get_bits_left(&gb) > 0) |
456 | av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n"); |
457 | |
458 | *got_frame_ptr = 1; |
459 | |
460 | return buf_size; |
461 | } |
462 | |
463 | static void g726_decode_flush(AVCodecContext *avctx) |
464 | { |
465 | G726Context *c = avctx->priv_data; |
466 | g726_reset(c); |
467 | } |
468 | #endif |
469 | |
470 | #if CONFIG_ADPCM_G726_DECODER |
471 | AVCodec ff_adpcm_g726_decoder = { |
472 | .name = "g726", |
473 | .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), |
474 | .type = AVMEDIA_TYPE_AUDIO, |
475 | .id = AV_CODEC_ID_ADPCM_G726, |
476 | .priv_data_size = sizeof(G726Context), |
477 | .init = g726_decode_init, |
478 | .decode = g726_decode_frame, |
479 | .flush = g726_decode_flush, |
480 | .capabilities = AV_CODEC_CAP_DR1, |
481 | }; |
482 | #endif |
483 | |
484 | #if CONFIG_ADPCM_G726LE_DECODER |
485 | AVCodec ff_adpcm_g726le_decoder = { |
486 | .name = "g726le", |
487 | .type = AVMEDIA_TYPE_AUDIO, |
488 | .id = AV_CODEC_ID_ADPCM_G726LE, |
489 | .priv_data_size = sizeof(G726Context), |
490 | .init = g726_decode_init, |
491 | .decode = g726_decode_frame, |
492 | .flush = g726_decode_flush, |
493 | .capabilities = AV_CODEC_CAP_DR1, |
494 | .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM little-endian"), |
495 | }; |
496 | #endif |
497 |