g726: treat sample rates other than 8kHz as unofficial.
[libav.git] / libavcodec / g726.c
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 Libav.
9 *
10 * Libav 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 * Libav 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 Libav; 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 #include "avcodec.h"
26 #include "get_bits.h"
27 #include "put_bits.h"
28
29 /**
30 * G.726 11bit float.
31 * G.726 Standard uses rather odd 11bit floating point arithmentic for
32 * numerous occasions. It's a mistery to me why they did it this way
33 * instead of simply using 32bit integer arithmetic.
34 */
35 typedef struct Float11 {
36 uint8_t sign; /**< 1bit sign */
37 uint8_t exp; /**< 4bit exponent */
38 uint8_t mant; /**< 6bit mantissa */
39 } Float11;
40
41 static inline Float11* i2f(int i, Float11* f)
42 {
43 f->sign = (i < 0);
44 if (f->sign)
45 i = -i;
46 f->exp = av_log2_16bit(i) + !!i;
47 f->mant = i? (i<<6) >> f->exp : 1<<5;
48 return f;
49 }
50
51 static inline int16_t mult(Float11* f1, Float11* f2)
52 {
53 int res, exp;
54
55 exp = f1->exp + f2->exp;
56 res = (((f1->mant * f2->mant) + 0x30) >> 4);
57 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
58 return (f1->sign ^ f2->sign) ? -res : res;
59 }
60
61 static inline int sgn(int value)
62 {
63 return (value < 0) ? -1 : 1;
64 }
65
66 typedef struct G726Tables {
67 const int* quant; /**< quantization table */
68 const int16_t* iquant; /**< inverse quantization table */
69 const int16_t* W; /**< special table #1 ;-) */
70 const uint8_t* F; /**< special table #2 */
71 } G726Tables;
72
73 typedef struct G726Context {
74 G726Tables tbls; /**< static tables needed for computation */
75
76 Float11 sr[2]; /**< prev. reconstructed samples */
77 Float11 dq[6]; /**< prev. difference */
78 int a[2]; /**< second order predictor coeffs */
79 int b[6]; /**< sixth order predictor coeffs */
80 int pk[2]; /**< signs of prev. 2 sez + dq */
81
82 int ap; /**< scale factor control */
83 int yu; /**< fast scale factor */
84 int yl; /**< slow scale factor */
85 int dms; /**< short average magnitude of F[i] */
86 int dml; /**< long average magnitude of F[i] */
87 int td; /**< tone detect */
88
89 int se; /**< estimated signal for the next iteration */
90 int sez; /**< estimated second order prediction */
91 int y; /**< quantizer scaling factor for the next iteration */
92 int code_size;
93 } G726Context;
94
95 static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
96 { 260, INT_MAX };
97 static const int16_t iquant_tbl16[] =
98 { 116, 365, 365, 116 };
99 static const int16_t W_tbl16[] =
100 { -22, 439, 439, -22 };
101 static const uint8_t F_tbl16[] =
102 { 0, 7, 7, 0 };
103
104 static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
105 { 7, 217, 330, INT_MAX };
106 static const int16_t iquant_tbl24[] =
107 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
108 static const int16_t W_tbl24[] =
109 { -4, 30, 137, 582, 582, 137, 30, -4 };
110 static const uint8_t F_tbl24[] =
111 { 0, 1, 2, 7, 7, 2, 1, 0 };
112
113 static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
114 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
115 static const int16_t iquant_tbl32[] =
116 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
117 425, 373, 323, 273, 213, 135, 4, INT16_MIN };
118 static const int16_t W_tbl32[] =
119 { -12, 18, 41, 64, 112, 198, 355, 1122,
120 1122, 355, 198, 112, 64, 41, 18, -12};
121 static const uint8_t F_tbl32[] =
122 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
123
124 static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
125 { -122, -16, 67, 138, 197, 249, 297, 338,
126 377, 412, 444, 474, 501, 527, 552, INT_MAX };
127 static const int16_t iquant_tbl40[] =
128 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318,
129 358, 395, 429, 459, 488, 514, 539, 566,
130 566, 539, 514, 488, 459, 429, 395, 358,
131 318, 274, 224, 169, 104, 28, -66, INT16_MIN };
132 static const int16_t W_tbl40[] =
133 { 14, 14, 24, 39, 40, 41, 58, 100,
134 141, 179, 219, 280, 358, 440, 529, 696,
135 696, 529, 440, 358, 280, 219, 179, 141,
136 100, 58, 41, 40, 39, 24, 14, 14 };
137 static const uint8_t F_tbl40[] =
138 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
139 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
140
141 static const G726Tables G726Tables_pool[] =
142 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
143 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
144 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
145 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
146
147
148 /**
149 * Para 4.2.2 page 18: Adaptive quantizer.
150 */
151 static inline uint8_t quant(G726Context* c, int d)
152 {
153 int sign, exp, i, dln;
154
155 sign = i = 0;
156 if (d < 0) {
157 sign = 1;
158 d = -d;
159 }
160 exp = av_log2_16bit(d);
161 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
162
163 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
164 ++i;
165
166 if (sign)
167 i = ~i;
168 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
169 i = 0xff;
170
171 return i;
172 }
173
174 /**
175 * Para 4.2.3 page 22: Inverse adaptive quantizer.
176 */
177 static inline int16_t inverse_quant(G726Context* c, int i)
178 {
179 int dql, dex, dqt;
180
181 dql = c->tbls.iquant[i] + (c->y >> 2);
182 dex = (dql>>7) & 0xf; /* 4bit exponent */
183 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
184 return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
185 }
186
187 static int16_t g726_decode(G726Context* c, int I)
188 {
189 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
190 Float11 f;
191 int I_sig= I >> (c->code_size - 1);
192
193 dq = inverse_quant(c, I);
194
195 /* Transition detect */
196 ylint = (c->yl >> 15);
197 ylfrac = (c->yl >> 10) & 0x1f;
198 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
199 tr= (c->td == 1 && dq > ((3*thr2)>>2));
200
201 if (I_sig) /* get the sign */
202 dq = -dq;
203 re_signal = c->se + dq;
204
205 /* Update second order predictor coefficient A2 and A1 */
206 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
207 dq0 = dq ? sgn(dq) : 0;
208 if (tr) {
209 c->a[0] = 0;
210 c->a[1] = 0;
211 for (i=0; i<6; i++)
212 c->b[i] = 0;
213 } else {
214 /* This is a bit crazy, but it really is +255 not +256 */
215 fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
216
217 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
218 c->a[1] = av_clip(c->a[1], -12288, 12288);
219 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
220 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
221
222 for (i=0; i<6; i++)
223 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
224 }
225
226 /* Update Dq and Sr and Pk */
227 c->pk[1] = c->pk[0];
228 c->pk[0] = pk0 ? pk0 : 1;
229 c->sr[1] = c->sr[0];
230 i2f(re_signal, &c->sr[0]);
231 for (i=5; i>0; i--)
232 c->dq[i] = c->dq[i-1];
233 i2f(dq, &c->dq[0]);
234 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
235
236 c->td = c->a[1] < -11776;
237
238 /* Update Ap */
239 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
240 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
241 if (tr)
242 c->ap = 256;
243 else {
244 c->ap += (-c->ap) >> 4;
245 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
246 c->ap += 0x20;
247 }
248
249 /* Update Yu and Yl */
250 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
251 c->yl += c->yu + ((-c->yl)>>6);
252
253 /* Next iteration for Y */
254 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
255 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
256
257 /* Next iteration for SE and SEZ */
258 c->se = 0;
259 for (i=0; i<6; i++)
260 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
261 c->sez = c->se >> 1;
262 for (i=0; i<2; i++)
263 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
264 c->se >>= 1;
265
266 return av_clip(re_signal << 2, -0xffff, 0xffff);
267 }
268
269 static av_cold int g726_reset(G726Context* c, int index)
270 {
271 int i;
272
273 c->tbls = G726Tables_pool[index];
274 for (i=0; i<2; i++) {
275 c->sr[i].mant = 1<<5;
276 c->pk[i] = 1;
277 }
278 for (i=0; i<6; i++) {
279 c->dq[i].mant = 1<<5;
280 }
281 c->yu = 544;
282 c->yl = 34816;
283
284 c->y = 544;
285
286 return 0;
287 }
288
289 #if CONFIG_ADPCM_G726_ENCODER
290 static int16_t g726_encode(G726Context* c, int16_t sig)
291 {
292 uint8_t i;
293
294 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
295 g726_decode(c, i);
296 return i;
297 }
298
299 /* Interfacing to the libavcodec */
300
301 static av_cold int g726_encode_init(AVCodecContext *avctx)
302 {
303 G726Context* c = avctx->priv_data;
304
305 if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL &&
306 avctx->sample_rate != 8000) {
307 av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not "
308 "allowed when the compliance level is higher than unofficial. "
309 "Resample or reduce the compliance level.\n");
310 return AVERROR(EINVAL);
311 }
312 if (avctx->sample_rate <= 0) {
313 av_log(avctx, AV_LOG_ERROR, "Samplerate is invalid\n");
314 return -1;
315 }
316
317 if(avctx->channels != 1){
318 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
319 return -1;
320 }
321
322 if (avctx->bit_rate % avctx->sample_rate) {
323 av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n");
324 return AVERROR(EINVAL);
325 }
326 c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
327 if (c->code_size < 2 || c->code_size > 5) {
328 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
329 return AVERROR(EINVAL);
330 }
331 avctx->bits_per_coded_sample = c->code_size;
332
333 g726_reset(c, c->code_size - 2);
334
335 avctx->coded_frame = avcodec_alloc_frame();
336 if (!avctx->coded_frame)
337 return AVERROR(ENOMEM);
338 avctx->coded_frame->key_frame = 1;
339
340 /* select a frame size that will end on a byte boundary and have a size of
341 approximately 1024 bytes */
342 avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
343
344 return 0;
345 }
346
347 static av_cold int g726_encode_close(AVCodecContext *avctx)
348 {
349 av_freep(&avctx->coded_frame);
350 return 0;
351 }
352
353 static int g726_encode_frame(AVCodecContext *avctx,
354 uint8_t *dst, int buf_size, void *data)
355 {
356 G726Context *c = avctx->priv_data;
357 const int16_t *samples = data;
358 PutBitContext pb;
359 int i;
360
361 init_put_bits(&pb, dst, 1024*1024);
362
363 for (i = 0; i < avctx->frame_size; i++)
364 put_bits(&pb, c->code_size, g726_encode(c, *samples++));
365
366 flush_put_bits(&pb);
367
368 return put_bits_count(&pb)>>3;
369 }
370 #endif
371
372 static av_cold int g726_decode_init(AVCodecContext *avctx)
373 {
374 G726Context* c = avctx->priv_data;
375
376 if (avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
377 avctx->sample_rate != 8000) {
378 av_log(avctx, AV_LOG_ERROR, "Only 8kHz sample rate is allowed when "
379 "the compliance level is strict. Reduce the compliance level "
380 "if you wish to decode the stream anyway.\n");
381 return AVERROR(EINVAL);
382 }
383
384 if(avctx->channels != 1){
385 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
386 return -1;
387 }
388
389 c->code_size = avctx->bits_per_coded_sample;
390 if (c->code_size < 2 || c->code_size > 5) {
391 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
392 return AVERROR(EINVAL);
393 }
394 g726_reset(c, c->code_size - 2);
395
396 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
397
398 return 0;
399 }
400
401 static int g726_decode_frame(AVCodecContext *avctx,
402 void *data, int *data_size,
403 AVPacket *avpkt)
404 {
405 const uint8_t *buf = avpkt->data;
406 int buf_size = avpkt->size;
407 G726Context *c = avctx->priv_data;
408 int16_t *samples = data;
409 GetBitContext gb;
410 int out_samples, out_size;
411
412 out_samples = buf_size * 8 / c->code_size;
413 out_size = out_samples * av_get_bytes_per_sample(avctx->sample_fmt);
414 if (*data_size < out_size) {
415 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
416 return AVERROR(EINVAL);
417 }
418
419 init_get_bits(&gb, buf, buf_size * 8);
420
421 while (out_samples--)
422 *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
423
424 if (get_bits_left(&gb) > 0)
425 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
426
427 *data_size = out_size;
428 return buf_size;
429 }
430
431 #if CONFIG_ADPCM_G726_ENCODER
432 AVCodec ff_adpcm_g726_encoder = {
433 .name = "g726",
434 .type = AVMEDIA_TYPE_AUDIO,
435 .id = CODEC_ID_ADPCM_G726,
436 .priv_data_size = sizeof(G726Context),
437 .init = g726_encode_init,
438 .encode = g726_encode_frame,
439 .close = g726_encode_close,
440 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
441 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
442 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
443 };
444 #endif
445
446 AVCodec ff_adpcm_g726_decoder = {
447 .name = "g726",
448 .type = AVMEDIA_TYPE_AUDIO,
449 .id = CODEC_ID_ADPCM_G726,
450 .priv_data_size = sizeof(G726Context),
451 .init = g726_decode_init,
452 .decode = g726_decode_frame,
453 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
454 };