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 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 #include "avcodec.h"
  26 #include "bitstream.h"
  27
  28 /**
  29  * G.726 11bit float.
  30  * G.726 Standard uses rather odd 11bit floating point arithmentic for
  31  * numerous occasions. It's a mistery to me why they did it this way
  32  * instead of simply using 32bit integer arithmetic.
  33  */
  34 typedef struct Float11 {
  35     int sign;   /**< 1bit sign */
  36     int exp;    /**< 4bit exponent */
  37     int mant;   /**< 6bit mantissa */
  38 } Float11;
  39
  40 static inline Float11* i2f(int16_t i, Float11* f)
  41 {
  42     f->sign = (i < 0);
  43     if (f->sign)
  44         i = -i;
  45     f->exp = av_log2_16bit(i) + !!i;
  46     f->mant = i? (i<<6) >> f->exp : 1<<5;
  47     return f;
  48 }
  49
  50 static inline int16_t mult(Float11* f1, Float11* f2)
  51 {
  52         int res, exp;
  53
  54         exp = f1->exp + f2->exp;
  55         res = (((f1->mant * f2->mant) + 0x30) >> 4) << 7;
  56         res = exp > 26 ? res << (exp - 26) : res >> (26 - exp);
  57         return (f1->sign ^ f2->sign) ? -res : res;
  58 }
  59
  60 static inline int sgn(int value)
  61 {
  62     return (value < 0) ? -1 : 1;
  63 }
  64
  65 typedef struct G726Tables {
  66     int  bits;          /**< bits per sample */
  67     int* quant;         /**< quantization table */
  68     int* iquant;        /**< inverse quantization table */
  69     int* W;             /**< special table #1 ;-) */
  70     int* 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 } G726Context;
  93
  94 static int quant_tbl16[] =                  /**< 16kbit/s 2bits per sample */
  95            { 260, INT_MAX };
  96 static int iquant_tbl16[] =
  97            { 116, 365, 365, 116 };
  98 static int W_tbl16[] =
  99            { -22, 439, 439, -22 };
 100 static int F_tbl16[] =
 101            { 0, 7, 7, 0 };
 102
 103 static int quant_tbl24[] =                  /**< 24kbit/s 3bits per sample */
 104            {  7, 217, 330, INT_MAX };
 105 static int iquant_tbl24[] =
 106            { INT_MIN, 135, 273, 373, 373, 273, 135, INT_MIN };
 107 static int W_tbl24[] =
 108            { -4,  30, 137, 582, 582, 137,  30, -4 };
 109 static int F_tbl24[] =
 110            { 0, 1, 2, 7, 7, 2, 1, 0 };
 111
 112 static int quant_tbl32[] =                  /**< 32kbit/s 4bits per sample */
 113            { -125,  79, 177, 245, 299, 348, 399, INT_MAX };
 114 static int iquant_tbl32[] =
 115            { INT_MIN,   4, 135, 213, 273, 323, 373, 425,
 116                  425, 373, 323, 273, 213, 135,   4, INT_MIN };
 117 static int W_tbl32[] =
 118            { -12,  18,  41,  64, 112, 198, 355, 1122,
 119             1122, 355, 198, 112,  64,  41,  18, -12};
 120 static int F_tbl32[] =
 121            { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
 122
 123 static int quant_tbl40[] =                  /**< 40kbit/s 5bits per sample */
 124            { -122, -16,  67, 138, 197, 249, 297, 338,
 125               377, 412, 444, 474, 501, 527, 552, INT_MAX };
 126 static int iquant_tbl40[] =
 127            { INT_MIN, -66,  28, 104, 169, 224, 274, 318,
 128                  358, 395, 429, 459, 488, 514, 539, 566,
 129                  566, 539, 514, 488, 459, 429, 395, 358,
 130                  318, 274, 224, 169, 104,  28, -66, INT_MIN };
 131 static int W_tbl40[] =
 132            {   14,  14,  24,  39,  40,  41,   58,  100,
 133               141, 179, 219, 280, 358, 440,  529,  696,
 134               696, 529, 440, 358, 280, 219,  179,  141,
 135               100,  58,  41,  40,  39,  24,   14,   14 };
 136 static int F_tbl40[] =
 137            { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
 138              6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
 139
 140 static G726Tables G726Tables_pool[] =
 141            {{ 2, quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
 142             { 3, quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
 143             { 4, quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
 144             { 5, quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
 145
 146
 147 /**
 148  * Para 4.2.2 page 18: Adaptive quantizer.
 149  */
 150 static inline uint8_t quant(G726Context* c, int d)
 151 {
 152     int sign, exp, i, dln;
 153
 154     sign = i = 0;
 155     if (d < 0) {
 156         sign = 1;
 157         d = -d;
 158     }
 159     exp = av_log2_16bit(d);
 160     dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
 161
 162     while (c->tbls->quant[i] < INT_MAX && c->tbls->quant[i] < dln)
 163         ++i;
 164
 165     if (sign)
 166         i = ~i;
 167     if (c->tbls->bits != 2 && i == 0) /* I'm not sure this is a good idea */
 168         i = 0xff;
 169
 170     return i;
 171 }
 172
 173 /**
 174  * Para 4.2.3 page 22: Inverse adaptive quantizer.
 175  */
 176 static inline int16_t inverse_quant(G726Context* c, int i)
 177 {
 178     int dql, dex, dqt;
 179
 180     dql = c->tbls->iquant[i] + (c->y >> 2);
 181     dex = (dql>>7) & 0xf;        /* 4bit exponent */
 182     dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
 183     return (dql < 0) ? 0 : ((dqt<<7) >> (14-dex));
 184 }
 185
 186 static inline int16_t g726_iterate(G726Context* c, int16_t I)
 187 {
 188     int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
 189     Float11 f;
 190
 191     dq = inverse_quant(c, I);
 192     if (I >> (c->tbls->bits - 1))  /* get the sign */
 193         dq = -dq;
 194     re_signal = c->se + dq;
 195
 196     /* Transition detect */
 197     ylint = (c->yl >> 15);
 198     ylfrac = (c->yl >> 10) & 0x1f;
 199     thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
 200     if (c->td == 1 && abs(dq) > ((thr2+(thr2>>1))>>1))
 201         tr = 1;
 202     else
 203         tr = 0;
 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 >> (c->tbls->bits - 1); /* Isn't it crazy ?!?! */
 235
 236     /* Update tone detect [I'm not sure 'tr == 0' is really needed] */
 237     c->td = (tr == 0 && c->a[1] < -11776);
 238
 239     /* Update Ap */
 240     c->dms += ((c->tbls->F[I]<<9) - c->dms) >> 5;
 241     c->dml += ((c->tbls->F[I]<<11) - c->dml) >> 7;
 242     if (tr)
 243         c->ap = 256;
 244     else if (c->y > 1535 && !c->td && (abs((c->dms << 2) - c->dml) < (c->dml >> 3)))
 245         c->ap += (-c->ap) >> 4;
 246     else
 247         c->ap += (0x200 - c->ap) >> 4;
 248
 249     /* Update Yu and Yl */
 250     c->yu = av_clip(c->y + (((c->tbls->W[I] << 5) - 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 int g726_reset(G726Context* c, int bit_rate)
 270 {
 271     int i;
 272
 273     c->tbls = &G726Tables_pool[bit_rate/8000 - 2];
 274     for (i=0; i<2; i++) {
 275         i2f(0, &c->sr[i]);
 276         c->a[i] = 0;
 277         c->pk[i] = 1;
 278     }
 279     for (i=0; i<6; i++) {
 280         i2f(0, &c->dq[i]);
 281         c->b[i] = 0;
 282     }
 283     c->ap = 0;
 284     c->dms = 0;
 285     c->dml = 0;
 286     c->yu = 544;
 287     c->yl = 34816;
 288     c->td = 0;
 289
 290     c->se = 0;
 291     c->sez = 0;
 292     c->y = 544;
 293
 294     return 0;
 295 }
 296
 297 static int16_t g726_decode(G726Context* c, int16_t i)
 298 {
 299     return g726_iterate(c, i);
 300 }
 301
 302 #ifdef CONFIG_ENCODERS
 303 static int16_t g726_encode(G726Context* c, int16_t sig)
 304 {
 305     uint8_t i;
 306
 307     i = quant(c, sig/4 - c->se) & ((1<<c->tbls->bits) - 1);
 308     g726_iterate(c, i);
 309     return i;
 310 }
 311 #endif
 312
 313 /* Interfacing to the libavcodec */
 314
 315 typedef struct AVG726Context {
 316     G726Context c;
 317     int bits_left;
 318     int bit_buffer;
 319     int code_size;
 320 } AVG726Context;
 321
 322 static int g726_init(AVCodecContext * avctx)
 323 {
 324     AVG726Context* c = (AVG726Context*)avctx->priv_data;
 325
 326     if (avctx->channels != 1 ||
 327         (avctx->bit_rate != 16000 && avctx->bit_rate != 24000 &&
 328          avctx->bit_rate != 32000 && avctx->bit_rate != 40000)) {
 329         av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
 330         return -1;
 331     }
 332     if (avctx->sample_rate != 8000 && avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL) {
 333         av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
 334         return -1;
 335     }
 336     g726_reset(&c->c, avctx->bit_rate);
 337     c->code_size = c->c.tbls->bits;
 338     c->bit_buffer = 0;
 339     c->bits_left = 0;
 340
 341     avctx->coded_frame = avcodec_alloc_frame();
 342     if (!avctx->coded_frame)
 343         return AVERROR(ENOMEM);
 344     avctx->coded_frame->key_frame = 1;
 345
 346     return 0;
 347 }
 348
 349 static int g726_close(AVCodecContext *avctx)
 350 {
 351     av_freep(&avctx->coded_frame);
 352     return 0;
 353 }
 354
 355 #ifdef CONFIG_ENCODERS
 356 static int g726_encode_frame(AVCodecContext *avctx,
 357                             uint8_t *dst, int buf_size, void *data)
 358 {
 359     AVG726Context *c = avctx->priv_data;
 360     short *samples = data;
 361     PutBitContext pb;
 362
 363     init_put_bits(&pb, dst, 1024*1024);
 364
 365     for (; buf_size; buf_size--)
 366         put_bits(&pb, c->code_size, g726_encode(&c->c, *samples++));
 367
 368     flush_put_bits(&pb);
 369
 370     return put_bits_count(&pb)>>3;
 371 }
 372 #endif
 373
 374 static int g726_decode_frame(AVCodecContext *avctx,
 375                              void *data, int *data_size,
 376                              uint8_t *buf, int buf_size)
 377 {
 378     AVG726Context *c = avctx->priv_data;
 379     short *samples = data;
 380     uint8_t code;
 381     uint8_t mask;
 382     GetBitContext gb;
 383
 384     if (!buf_size)
 385         goto out;
 386
 387     mask = (1<<c->code_size) - 1;
 388     init_get_bits(&gb, buf, buf_size * 8);
 389     if (c->bits_left) {
 390         int s = c->code_size - c->bits_left;;
 391         code = (c->bit_buffer << s) | get_bits(&gb, s);
 392         *samples++ = g726_decode(&c->c, code & mask);
 393     }
 394
 395     while (get_bits_count(&gb) + c->code_size <= buf_size*8)
 396         *samples++ = g726_decode(&c->c, get_bits(&gb, c->code_size) & mask);
 397
 398     c->bits_left = buf_size*8 - get_bits_count(&gb);
 399     c->bit_buffer = get_bits(&gb, c->bits_left);
 400
 401 out:
 402     *data_size = (uint8_t*)samples - (uint8_t*)data;
 403     return buf_size;
 404 }
 405
 406 #ifdef CONFIG_ENCODERS
 407 AVCodec adpcm_g726_encoder = {
 408     "g726",
 409     CODEC_TYPE_AUDIO,
 410     CODEC_ID_ADPCM_G726,
 411     sizeof(AVG726Context),
 412     g726_init,
 413     g726_encode_frame,
 414     g726_close,
 415     NULL,
 416 };
 417 #endif //CONFIG_ENCODERS
 418
 419 AVCodec adpcm_g726_decoder = {
 420     "g726",
 421     CODEC_TYPE_AUDIO,
 422     CODEC_ID_ADPCM_G726,
 423     sizeof(AVG726Context),
 424     g726_init,
 425     NULL,
 426     g726_close,
 427     g726_decode_frame,
 428 };