## ffmpeg / libavcodec / g726.c @ 58c2182d

History | View | Annotate | Download (11.9 KB)

1 | e5966052 | Roman Shaposhnik | ```
/*
``` |
---|---|---|---|

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 library is free software; you can redistribute it and/or
``` |
||

9 | ```
* modify it under the terms of the GNU Lesser General Public
``` |
||

10 | ```
* License as published by the Free Software Foundation; either
``` |
||

11 | ```
* version 2 of the License, or (at your option) any later version.
``` |
||

12 | ```
*
``` |
||

13 | ```
* This library is distributed in the hope that it will be useful,
``` |
||

14 | ```
* but WITHOUT ANY WARRANTY; without even the implied warranty of
``` |
||

15 | ```
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
``` |
||

16 | ```
* Lesser General Public License for more details.
``` |
||

17 | ```
*
``` |
||

18 | ```
* You should have received a copy of the GNU Lesser General Public
``` |
||

19 | ```
* License along with this library; if not, write to the Free Software
``` |
||

20 | ```
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
``` |
||

21 | ```
*/
``` |
||

22 | #include <limits.h> |
||

23 | #include "avcodec.h" |
||

24 | #include "common.h" |
||

25 | |||

26 | ```
/*
``` |
||

27 | ```
* G.726 Standard uses rather odd 11bit floating point arithmentic for
``` |
||

28 | ```
* numerous occasions. It's a mistery to me why they did it this way
``` |
||

29 | ```
* instead of simply using 32bit integer arithmetic.
``` |
||

30 | ```
*/
``` |
||

31 | typedef struct Float11 { |
||

32 | int sign; /* 1bit sign */ |
||

33 | int exp; /* 4bit exponent */ |
||

34 | int mant; /* 6bit mantissa */ |
||

35 | } Float11; |
||

36 | |||

37 | static inline Float11* i2f(int16_t i, Float11* f) |
||

38 | { |
||

39 | ```
f->sign = (i < 0);
``` |
||

40 | ```
if (f->sign)
``` |
||

41 | i = -i; |
||

42 | f->exp = av_log2_16bit(i) + !!i; |
||

43 | ```
f->mant = i? (i<<6) >> f->exp :
``` |
||

44 | 1<<5; |
||

45 | ```
return f;
``` |
||

46 | } |
||

47 | |||

48 | static inline int16_t mult(Float11* f1, Float11* f2) |
||

49 | { |
||

50 | ```
int res, exp;
``` |
||

51 | |||

52 | exp = f1->exp + f2->exp; |
||

53 | res = (((f1->mant * f2->mant) + 0x30) >> 4) << 7; |
||

54 | res = exp > 26 ? res << (exp - 26) : res >> (26 - exp); |
||

55 | ```
return (f1->sign ^ f2->sign) ? -res : res;
``` |
||

56 | } |
||

57 | |||

58 | static inline int sgn(int value) |
||

59 | { |
||

60 | return (value < 0) ? -1 : 1; |
||

61 | } |
||

62 | |||

63 | typedef struct G726Tables { |
||

64 | int bits; /* bits per sample */ |
||

65 | int* quant; /* quantization table */ |
||

66 | int* iquant; /* inverse quantization table */ |
||

67 | int* W; /* special table #1 ;-) */ |
||

68 | int* F; /* special table #2 */ |
||

69 | } G726Tables; |
||

70 | |||

71 | typedef struct G726Context { |
||

72 | ```
G726Tables* tbls; /* static tables needed for computation */
``` |
||

73 | |||

74 | Float11 sr[2]; /* prev. reconstructed samples */ |
||

75 | Float11 dq[6]; /* prev. difference */ |
||

76 | int a[2]; /* second order predictor coeffs */ |
||

77 | int b[6]; /* sixth order predictor coeffs */ |
||

78 | int pk[2]; /* signs of prev. 2 sez + dq */ |
||

79 | |||

80 | int ap; /* scale factor control */ |
||

81 | int yu; /* fast scale factor */ |
||

82 | int yl; /* slow scale factor */ |
||

83 | int dms; /* short average magnitude of F[i] */ |
||

84 | int dml; /* long average magnitude of F[i] */ |
||

85 | int td; /* tone detect */ |
||

86 | |||

87 | int se; /* estimated signal for the next iteration */ |
||

88 | int sez; /* estimated second order prediction */ |
||

89 | int y; /* quantizer scaling factor for the next iteration */ |
||

90 | } G726Context; |
||

91 | |||

92 | static int quant_tbl16[] = /* 16kbit/s 2bits per sample */ |
||

93 | ```
{ 260, INT_MAX };
``` |
||

94 | static int iquant_tbl16[] = |
||

95 | { 116, 365, 365, 116 }; |
||

96 | static int W_tbl16[] = |
||

97 | { -22, 439, 439, -22 }; |
||

98 | static int F_tbl16[] = |
||

99 | { 0, 7, 7, 0 }; |
||

100 | |||

101 | static int quant_tbl24[] = /* 24kbit/s 3bits per sample */ |
||

102 | { 7, 217, 330, INT_MAX }; |
||

103 | static int iquant_tbl24[] = |
||

104 | { INT_MIN, 135, 273, 373, 373, 273, 135, INT_MIN }; |
||

105 | static int W_tbl24[] = |
||

106 | { -4, 30, 137, 582, 582, 137, 30, -4 }; |
||

107 | static int F_tbl24[] = |
||

108 | { 0, 1, 2, 7, 7, 2, 1, 0 }; |
||

109 | |||

110 | static int quant_tbl32[] = /* 32kbit/s 4bits per sample */ |
||

111 | { -125, 79, 177, 245, 299, 348, 399, INT_MAX }; |
||

112 | static int iquant_tbl32[] = |
||

113 | { INT_MIN, 4, 135, 213, 273, 323, 373, 425, |
||

114 | 425, 373, 323, 273, 213, 135, 4, INT_MIN }; |
||

115 | static int W_tbl32[] = |
||

116 | { -12, 18, 41, 64, 112, 198, 355, 1122, |
||

117 | 1122, 355, 198, 112, 64, 41, 18, -12}; |
||

118 | static int F_tbl32[] = |
||

119 | { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 }; |
||

120 | |||

121 | static int quant_tbl40[] = /* 40kbit/s 5bits per sample */ |
||

122 | { -122, -16, 67, 138, 197, 249, 297, 338, |
||

123 | 377, 412, 444, 474, 501, 527, 552, INT_MAX }; |
||

124 | static int iquant_tbl40[] = |
||

125 | { INT_MIN, -66, 28, 104, 169, 224, 274, 318, |
||

126 | 358, 395, 429, 459, 488, 514, 539, 566, |
||

127 | 566, 539, 514, 488, 459, 429, 395, 358, |
||

128 | 318, 274, 224, 169, 104, 28, -66, INT_MIN }; |
||

129 | static int W_tbl40[] = |
||

130 | { 14, 14, 24, 39, 40, 41, 58, 100, |
||

131 | 141, 179, 219, 280, 358, 440, 529, 696, |
||

132 | 696, 529, 440, 358, 280, 219, 179, 141, |
||

133 | 100, 58, 41, 40, 39, 24, 14, 14 }; |
||

134 | static int F_tbl40[] = |
||

135 | { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6, |
||

136 | 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 }; |
||

137 | |||

138 | ```
static G726Tables G726Tables_pool[] =
``` |
||

139 | ```
{{ 2, quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
``` |
||

140 | ```
{ 3, quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
``` |
||

141 | ```
{ 4, quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
``` |
||

142 | ```
{ 5, quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
``` |
||

143 | |||

144 | |||

145 | ```
/*
``` |
||

146 | ```
* Para 4.2.2 page 18: Adaptive quantizer.
``` |
||

147 | ```
*/
``` |
||

148 | static inline uint8_t quant(G726Context* c, int d) |
||

149 | { |
||

150 | ```
int sign, exp, i, dln;
``` |
||

151 | |||

152 | ```
sign = i = 0;
``` |
||

153 | if (d < 0) { |
||

154 | ```
sign = 1;
``` |
||

155 | d = -d; |
||

156 | } |
||

157 | exp = av_log2_16bit(d); |
||

158 | dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2); |
||

159 | |||

160 | ```
while (c->tbls->quant[i] < INT_MAX && c->tbls->quant[i] < dln)
``` |
||

161 | ++i; |
||

162 | |||

163 | ```
if (sign)
``` |
||

164 | i = ~i; |
||

165 | if (c->tbls->bits != 2 && i == 0) /* I'm not sure this is a good idea */ |
||

166 | ```
i = 0xff;
``` |
||

167 | |||

168 | ```
return i;
``` |
||

169 | } |
||

170 | |||

171 | ```
/*
``` |
||

172 | ```
* Para 4.2.3 page 22: Inverse adaptive quantizer.
``` |
||

173 | ```
*/
``` |
||

174 | static inline int16_t inverse_quant(G726Context* c, int i) |
||

175 | { |
||

176 | ```
int dql, dex, dqt;
``` |
||

177 | |||

178 | ```
dql = c->tbls->iquant[i] + (c->y >> 2);
``` |
||

179 | dex = (dql>>7) & 0xf; /* 4bit exponent */ |
||

180 | dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */ |
||

181 | return (dql < 0) ? 0 : ((dqt<<7) >> (14-dex)); |
||

182 | } |
||

183 | |||

184 | static inline int16_t g726_iterate(G726Context* c, int16_t I) |
||

185 | { |
||

186 | ```
int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
``` |
||

187 | Float11 f; |
||

188 | |||

189 | dq = inverse_quant(c, I); |
||

190 | if (I >> (c->tbls->bits - 1)) /* get the sign */ |
||

191 | dq = -dq; |
||

192 | re_signal = c->se + dq; |
||

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 | if (c->td == 1 && abs(dq) > ((thr2+(thr2>>1))>>1)) |
||

199 | ```
tr = 1;
``` |
||

200 | ```
else
``` |
||

201 | ```
tr = 0;
``` |
||

202 | |||

203 | ```
/* Update second order predictor coefficient A2 and A1 */
``` |
||

204 | ```
pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
``` |
||

205 | ```
dq0 = dq ? sgn(dq) : 0;
``` |
||

206 | ```
if (tr) {
``` |
||

207 | c->a[0] = 0; |
||

208 | c->a[1] = 0; |
||

209 | for (i=0; i<6; i++) |
||

210 | ```
c->b[i] = 0;
``` |
||

211 | ```
} else {
``` |
||

212 | ```
/* This is a bit crazy, but it really is +255 not +256 */
``` |
||

213 | eb5b0422 | Michael Niedermayer | fa1 = clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255); |

214 | e5966052 | Roman Shaposhnik | |

215 | c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7); |
||

216 | eb5b0422 | Michael Niedermayer | c->a[1] = clip(c->a[1], -12288, 12288); |

217 | e5966052 | Roman Shaposhnik | c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8); |

218 | eb5b0422 | Michael Niedermayer | c->a[0] = clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]); |

219 | e5966052 | Roman Shaposhnik | |

220 | for (i=0; i<6; i++) |
||

221 | c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8); |
||

222 | } |
||

223 | |||

224 | ```
/* Update Dq and Sr and Pk */
``` |
||

225 | c->pk[1] = c->pk[0]; |
||

226 | c->pk[0] = pk0 ? pk0 : 1; |
||

227 | c->sr[1] = c->sr[0]; |
||

228 | ```
i2f(re_signal, &c->sr[0]);
``` |
||

229 | for (i=5; i>0; i--) |
||

230 | ```
c->dq[i] = c->dq[i-1];
``` |
||

231 | ```
i2f(dq, &c->dq[0]);
``` |
||

232 | c->dq[0].sign = I >> (c->tbls->bits - 1); /* Isn't it crazy ?!?! */ |
||

233 | |||

234 | ```
/* Update tone detect [I'm not sure 'tr == 0' is really needed] */
``` |
||

235 | c->td = (tr == 0 && c->a[1] < -11776); |
||

236 | |||

237 | ```
/* Update Ap */
``` |
||

238 | c->dms += ((c->tbls->F[I]<<9) - c->dms) >> 5; |
||

239 | c->dml += ((c->tbls->F[I]<<11) - c->dml) >> 7; |
||

240 | ```
if (tr)
``` |
||

241 | ```
c->ap = 256;
``` |
||

242 | else if (c->y > 1535 && !c->td && (abs((c->dms << 2) - c->dml) < (c->dml >> 3))) |
||

243 | ```
c->ap += (-c->ap) >> 4;
``` |
||

244 | ```
else
``` |
||

245 | c->ap += (0x200 - c->ap) >> 4; |
||

246 | |||

247 | ```
/* Update Yu and Yl */
``` |
||

248 | eb5b0422 | Michael Niedermayer | c->yu = clip(c->y + (((c->tbls->W[I] << 5) - c->y) >> 5), 544, 5120); |

249 | e5966052 | Roman Shaposhnik | ```
c->yl += c->yu + ((-c->yl)>>6);
``` |

250 | |||

251 | ```
/* Next iteration for Y */
``` |
||

252 | al = (c->ap >= 256) ? 1<<6 : c->ap >> 2; |
||

253 | c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6; |
||

254 | |||

255 | ```
/* Next iteration for SE and SEZ */
``` |
||

256 | ```
c->se = 0;
``` |
||

257 | for (i=0; i<6; i++) |
||

258 | ```
c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
``` |
||

259 | ```
c->sez = c->se >> 1;
``` |
||

260 | for (i=0; i<2; i++) |
||

261 | ```
c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
``` |
||

262 | ```
c->se >>= 1;
``` |
||

263 | |||

264 | eb5b0422 | Michael Niedermayer | return clip(re_signal << 2, -0xffff, 0xffff); |

265 | e5966052 | Roman Shaposhnik | } |

266 | |||

267 | static int g726_reset(G726Context* c, int bit_rate) |
||

268 | { |
||

269 | ```
int i;
``` |
||

270 | |||

271 | c->tbls = &G726Tables_pool[bit_rate/8000 - 2]; |
||

272 | for (i=0; i<2; i++) { |
||

273 | ```
i2f(0, &c->sr[i]);
``` |
||

274 | ```
c->a[i] = 0;
``` |
||

275 | ```
c->pk[i] = 1;
``` |
||

276 | } |
||

277 | for (i=0; i<6; i++) { |
||

278 | ```
i2f(0, &c->dq[i]);
``` |
||

279 | ```
c->b[i] = 0;
``` |
||

280 | } |
||

281 | ```
c->ap = 0;
``` |
||

282 | ```
c->dms = 0;
``` |
||

283 | ```
c->dml = 0;
``` |
||

284 | ```
c->yu = 544;
``` |
||

285 | ```
c->yl = 34816;
``` |
||

286 | ```
c->td = 0;
``` |
||

287 | |||

288 | ```
c->se = 0;
``` |
||

289 | ```
c->sez = 0;
``` |
||

290 | ```
c->y = 544;
``` |
||

291 | |||

292 | return 0; |
||

293 | } |
||

294 | |||

295 | ```
static int16_t g726_decode(G726Context* c, int16_t i)
``` |
||

296 | { |
||

297 | ```
return g726_iterate(c, i);
``` |
||

298 | } |
||

299 | |||

300 | ```
static int16_t g726_encode(G726Context* c, int16_t sig)
``` |
||

301 | { |
||

302 | uint8_t i; |
||

303 | |||

304 | i = quant(c, sig/4 - c->se) & ((1<<c->tbls->bits) - 1); |
||

305 | g726_iterate(c, i); |
||

306 | ```
return i;
``` |
||

307 | } |
||

308 | |||

309 | ```
/* Interfacing to the libavcodec */
``` |
||

310 | |||

311 | typedef struct AVG726Context { |
||

312 | G726Context c; |
||

313 | ```
int bits_left;
``` |
||

314 | ```
int bit_buffer;
``` |
||

315 | ```
int code_size;
``` |
||

316 | } AVG726Context; |
||

317 | |||

318 | static int g726_init(AVCodecContext * avctx) |
||

319 | { |
||

320 | AVG726Context* c = (AVG726Context*)avctx->priv_data; |
||

321 | |||

322 | eb5b0422 | Michael Niedermayer | if (avctx->channels != 1 || |

323 | e5966052 | Roman Shaposhnik | (avctx->bit_rate != 16000 && avctx->bit_rate != 24000 && |

324 | avctx->bit_rate != 32000 && avctx->bit_rate != 40000)) { |
||

325 | ```
av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
``` |
||

326 | return -1; |
||

327 | } |
||

328 | eb5b0422 | Michael Niedermayer | if (avctx->sample_rate != 8000 && avctx->strict_std_compliance>=0) { |

329 | ```
av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
``` |
||

330 | return -1; |
||

331 | } |
||

332 | e5966052 | Roman Shaposhnik | g726_reset(&c->c, avctx->bit_rate); |

333 | c->code_size = c->c.tbls->bits; |
||

334 | ```
c->bit_buffer = 0;
``` |
||

335 | ```
c->bits_left = 0;
``` |
||

336 | |||

337 | 826ca104 | Roman Shaposhnik | avctx->coded_frame = avcodec_alloc_frame(); |

338 | ```
if (!avctx->coded_frame)
``` |
||

339 | ```
return -ENOMEM;
``` |
||

340 | ```
avctx->coded_frame->key_frame = 1;
``` |
||

341 | |||

342 | return 0; |
||

343 | } |
||

344 | |||

345 | static int g726_close(AVCodecContext *avctx) |
||

346 | { |
||

347 | av_freep(&avctx->coded_frame); |
||

348 | e5966052 | Roman Shaposhnik | return 0; |

349 | } |
||

350 | |||

351 | static int g726_encode_frame(AVCodecContext *avctx, |
||

352 | uint8_t *dst, int buf_size, void *data) |
||

353 | { |
||

354 | AVG726Context *c = avctx->priv_data; |
||

355 | ```
short *samples = data;
``` |
||

356 | PutBitContext pb; |
||

357 | |||

358 | init_put_bits(&pb, dst, 1024*1024); |
||

359 | |||

360 | ```
for (; buf_size; buf_size--)
``` |
||

361 | put_bits(&pb, c->code_size, g726_encode(&c->c, *samples++)); |
||

362 | |||

363 | flush_put_bits(&pb); |
||

364 | |||

365 | return put_bits_count(&pb)>>3; |
||

366 | } |
||

367 | |||

368 | static int g726_decode_frame(AVCodecContext *avctx, |
||

369 | void *data, int *data_size, |
||

370 | ```
uint8_t *buf, int buf_size)
``` |
||

371 | { |
||

372 | AVG726Context *c = avctx->priv_data; |
||

373 | ```
short *samples = data;
``` |
||

374 | uint8_t code; |
||

375 | uint8_t mask; |
||

376 | GetBitContext gb; |
||

377 | |||

378 | ```
if (!buf_size)
``` |
||

379 | ```
goto out;
``` |
||

380 | |||

381 | mask = (1<<c->code_size) - 1; |
||

382 | ```
init_get_bits(&gb, buf, buf_size * 8);
``` |
||

383 | ```
if (c->bits_left) {
``` |
||

384 | ```
int s = c->code_size - c->bits_left;;
``` |
||

385 | code = (c->bit_buffer << s) | get_bits(&gb, s); |
||

386 | *samples++ = g726_decode(&c->c, code & mask); |
||

387 | } |
||

388 | |||

389 | while (get_bits_count(&gb) + c->code_size <= buf_size*8) |
||

390 | *samples++ = g726_decode(&c->c, get_bits(&gb, c->code_size) & mask); |
||

391 | |||

392 | ```
c->bits_left = buf_size*8 - get_bits_count(&gb);
``` |
||

393 | c->bit_buffer = get_bits(&gb, c->bits_left); |
||

394 | |||

395 | ```
out:
``` |
||

396 | *data_size = (uint8_t*)samples - (uint8_t*)data; |
||

397 | ```
return buf_size;
``` |
||

398 | } |
||

399 | |||

400 | ```
#ifdef CONFIG_ENCODERS
``` |
||

401 | AVCodec adpcm_g726_encoder = { |
||

402 | ```
"g726",
``` |
||

403 | CODEC_TYPE_AUDIO, |
||

404 | CODEC_ID_ADPCM_G726, |
||

405 | ```
sizeof(AVG726Context),
``` |
||

406 | g726_init, |
||

407 | g726_encode_frame, |
||

408 | 826ca104 | Roman Shaposhnik | g726_close, |

409 | e5966052 | Roman Shaposhnik | ```
NULL,
``` |

410 | }; |
||

411 | #endif //CONFIG_ENCODERS |
||

412 | |||

413 | AVCodec adpcm_g726_decoder = { |
||

414 | ```
"g726",
``` |
||

415 | CODEC_TYPE_AUDIO, |
||

416 | CODEC_ID_ADPCM_G726, |
||

417 | ```
sizeof(AVG726Context),
``` |
||

418 | g726_init, |
||

419 | ```
NULL,
``` |
||

420 | 826ca104 | Roman Shaposhnik | g726_close, |

421 | e5966052 | Roman Shaposhnik | g726_decode_frame, |

422 | }; |