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     uint8_t sign;   /**< 1bit sign */
  36     uint8_t exp;    /**< 4bit exponent */
  37     uint8_t mant;   /**< 6bit mantissa */
  38 } Float11;
  39
  40 static inline Float11* i2f(int 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);
  56         res = exp > 19 ? res << (exp - 19) : res >> (19 - 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     const int* quant;         /**< quantization table */
  67     const int16_t* iquant;    /**< inverse quantization table */
  68     const int16_t* W;         /**< special table #1 ;-) */
  69     const uint8_t* F;         /**< special table #2 */
  70 } G726Tables;
  71
  72 typedef struct G726Context {
  73     G726Tables tbls;    /**< static tables needed for computation */
  74
  75     Float11 sr[2];      /**< prev. reconstructed samples */
  76     Float11 dq[6];      /**< prev. difference */
  77     int a[2];           /**< second order predictor coeffs */
  78     int b[6];           /**< sixth order predictor coeffs */
  79     int pk[2];          /**< signs of prev. 2 sez + dq */
  80
  81     int ap;             /**< scale factor control */
  82     int yu;             /**< fast scale factor */
  83     int yl;             /**< slow scale factor */
  84     int dms;            /**< short average magnitude of F[i] */
  85     int dml;            /**< long average magnitude of F[i] */
  86     int td;             /**< tone detect */
  87
  88     int se;             /**< estimated signal for the next iteration */
  89     int sez;            /**< estimated second order prediction */
  90     int y;              /**< quantizer scaling factor for the next iteration */
  91     int code_size;
  92 } G726Context;
  93
  94 static const int quant_tbl16[] =                  /**< 16kbit/s 2bits per sample */
  95            { 260, INT_MAX };
  96 static const int16_t iquant_tbl16[] =
  97            { 116, 365, 365, 116 };
  98 static const int16_t W_tbl16[] =
  99            { -22, 439, 439, -22 };
 100 static const uint8_t F_tbl16[] =
 101            { 0, 7, 7, 0 };
 102
 103 static const int quant_tbl24[] =                  /**< 24kbit/s 3bits per sample */
 104            {  7, 217, 330, INT_MAX };
 105 static const int16_t iquant_tbl24[] =
 106            { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
 107 static const int16_t W_tbl24[] =
 108            { -4,  30, 137, 582, 582, 137,  30, -4 };
 109 static const uint8_t F_tbl24[] =
 110            { 0, 1, 2, 7, 7, 2, 1, 0 };
 111
 112 static const int quant_tbl32[] =                  /**< 32kbit/s 4bits per sample */
 113            { -125,  79, 177, 245, 299, 348, 399, INT_MAX };
 114 static const int16_t iquant_tbl32[] =
 115          { INT16_MIN,   4, 135, 213, 273, 323, 373, 425,
 116                  425, 373, 323, 273, 213, 135,   4, INT16_MIN };
 117 static const int16_t W_tbl32[] =
 118            { -12,  18,  41,  64, 112, 198, 355, 1122,
 119             1122, 355, 198, 112,  64,  41,  18, -12};
 120 static const uint8_t F_tbl32[] =
 121            { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
 122
 123 static const 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 const int16_t iquant_tbl40[] =
 127          { INT16_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, INT16_MIN };
 131 static const int16_t 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 const uint8_t 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 const G726Tables G726Tables_pool[] =
 141            {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
 142             { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
 143             { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
 144             { 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->code_size != 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<<dex) >> 7);
 184 }
 185
 186 static int16_t g726_decode(G726Context* c, int I)
 187 {
 188     int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
 189     Float11 f;
 190     int I_sig= I >> (c->code_size - 1);
 191
 192     dq = inverse_quant(c, I);
 193
 194     /* Transition detect */
 195     ylint = (c->yl >> 15);
 196     ylfrac = (c->yl >> 10) & 0x1f;
 197     thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
 198     tr= (c->td == 1 && dq > ((3*thr2)>>2));
 199
 200     if (I_sig)  /* get the sign */
 201         dq = -dq;
 202     re_signal = c->se + dq;
 203
 204     /* Update second order predictor coefficient A2 and A1 */
 205     pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
 206     dq0 = dq ? sgn(dq) : 0;
 207     if (tr) {
 208         c->a[0] = 0;
 209         c->a[1] = 0;
 210         for (i=0; i<6; i++)
 211             c->b[i] = 0;
 212     } else {
 213         /* This is a bit crazy, but it really is +255 not +256 */
 214         fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
 215
 216         c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
 217         c->a[1] = av_clip(c->a[1], -12288, 12288);
 218         c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
 219         c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
 220
 221         for (i=0; i<6; i++)
 222             c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
 223     }
 224
 225     /* Update Dq and Sr and Pk */
 226     c->pk[1] = c->pk[0];
 227     c->pk[0] = pk0 ? pk0 : 1;
 228     c->sr[1] = c->sr[0];
 229     i2f(re_signal, &c->sr[0]);
 230     for (i=5; i>0; i--)
 231         c->dq[i] = c->dq[i-1];
 232     i2f(dq, &c->dq[0]);
 233     c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
 234
 235     c->td = c->a[1] < -11776;
 236
 237     /* Update Ap */
 238     c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
 239     c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
 240     if (tr)
 241         c->ap = 256;
 242     else {
 243         c->ap += (-c->ap) >> 4;
 244         if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
 245             c->ap += 0x20;
 246     }
 247
 248     /* Update Yu and Yl */
 249     c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
 250     c->yl += c->yu + ((-c->yl)>>6);
 251
 252     /* Next iteration for Y */
 253     al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
 254     c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
 255
 256     /* Next iteration for SE and SEZ */
 257     c->se = 0;
 258     for (i=0; i<6; i++)
 259         c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
 260     c->sez = c->se >> 1;
 261     for (i=0; i<2; i++)
 262         c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
 263     c->se >>= 1;
 264
 265     return av_clip(re_signal << 2, -0xffff, 0xffff);
 266 }
 267
 268 static av_cold int g726_reset(G726Context* c, int index)
 269 {
 270     int i;
 271
 272     c->tbls = G726Tables_pool[index];
 273     for (i=0; i<2; i++) {
 274         c->sr[i].mant = 1<<5;
 275         c->pk[i] = 1;
 276     }
 277     for (i=0; i<6; i++) {
 278         c->dq[i].mant = 1<<5;
 279     }
 280     c->yu = 544;
 281     c->yl = 34816;
 282
 283     c->y = 544;
 284
 285     return 0;
 286 }
 287
 288 #if CONFIG_ADPCM_G726_ENCODER
 289 static int16_t g726_encode(G726Context* c, int16_t sig)
 290 {
 291     uint8_t i;
 292
 293     i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
 294     g726_decode(c, i);
 295     return i;
 296 }
 297 #endif
 298
 299 /* Interfacing to the libavcodec */
 300
 301 static av_cold int g726_init(AVCodecContext * avctx)
 302 {
 303     G726Context* c = avctx->priv_data;
 304     unsigned int index;
 305
 306     if (avctx->sample_rate <= 0) {
 307         av_log(avctx, AV_LOG_ERROR, "Samplerate is invalid\n");
 308         return -1;
 309     }
 310
 311     index = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate - 2;
 312
 313     if (avctx->bit_rate % avctx->sample_rate && avctx->codec->encode) {
 314         av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n");
 315         return -1;
 316     }
 317     if(avctx->channels != 1){
 318         av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
 319         return -1;
 320     }
 321     if(index>3){
 322         av_log(avctx, AV_LOG_ERROR, "Unsupported number of bits %d\n", index+2);
 323         return -1;
 324     }
 325     g726_reset(c, index);
 326     c->code_size = index+2;
 327
 328     avctx->coded_frame = avcodec_alloc_frame();
 329     if (!avctx->coded_frame)
 330         return AVERROR(ENOMEM);
 331     avctx->coded_frame->key_frame = 1;
 332
 333     if (avctx->codec->decode)
 334         avctx->sample_fmt = SAMPLE_FMT_S16;
 335
 336     return 0;
 337 }
 338
 339 static av_cold int g726_close(AVCodecContext *avctx)
 340 {
 341     av_freep(&avctx->coded_frame);
 342     return 0;
 343 }
 344
 345 #if CONFIG_ADPCM_G726_ENCODER
 346 static int g726_encode_frame(AVCodecContext *avctx,
 347                             uint8_t *dst, int buf_size, void *data)
 348 {
 349     G726Context *c = avctx->priv_data;
 350     short *samples = data;
 351     PutBitContext pb;
 352
 353     init_put_bits(&pb, dst, 1024*1024);
 354
 355     for (; buf_size; buf_size--)
 356         put_bits(&pb, c->code_size, g726_encode(c, *samples++));
 357
 358     flush_put_bits(&pb);
 359
 360     return put_bits_count(&pb)>>3;
 361 }
 362 #endif
 363
 364 static int g726_decode_frame(AVCodecContext *avctx,
 365                              void *data, int *data_size,
 366                              const uint8_t *buf, int buf_size)
 367 {
 368     G726Context *c = avctx->priv_data;
 369     short *samples = data;
 370     GetBitContext gb;
 371
 372     init_get_bits(&gb, buf, buf_size * 8);
 373
 374     while (get_bits_count(&gb) + c->code_size <= buf_size*8)
 375         *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
 376
 377     if(buf_size*8 != get_bits_count(&gb))
 378         av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
 379
 380     *data_size = (uint8_t*)samples - (uint8_t*)data;
 381     return buf_size;
 382 }
 383
 384 #if CONFIG_ADPCM_G726_ENCODER
 385 AVCodec adpcm_g726_encoder = {
 386     "g726",
 387     CODEC_TYPE_AUDIO,
 388     CODEC_ID_ADPCM_G726,
 389     sizeof(G726Context),
 390     g726_init,
 391     g726_encode_frame,
 392     g726_close,
 393     NULL,
 394     .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
 395     .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
 396 };
 397 #endif
 398
 399 AVCodec adpcm_g726_decoder = {
 400     "g726",
 401     CODEC_TYPE_AUDIO,
 402     CODEC_ID_ADPCM_G726,
 403     sizeof(G726Context),
 404     g726_init,
 405     NULL,
 406     g726_close,
 407     g726_decode_frame,
 408     .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
 409 };