Mercurial > libavcodec.hg
annotate qcelpdec.c @ 8240:d3d0d9cc0e50 libavcodec
Commit last ok'ed parts of QCELP decoder and enable it.
patch by Kenan Gillet, kenan.gillet gmail com
author | vitor |
---|---|
date | Tue, 02 Dec 2008 16:48:05 +0000 |
parents | 93ee77578391 |
children | 75ae6859ac73 |
rev | line source |
---|---|
8096 | 1 /* |
2 * QCELP decoder | |
3 * Copyright (c) 2007 Reynaldo H. Verdejo Pinochet | |
4 * | |
5 * This file is part of FFmpeg. | |
6 * | |
7 * FFmpeg is free software; you can redistribute it and/or | |
8 * modify it under the terms of the GNU Lesser General Public | |
9 * License as published by the Free Software Foundation; either | |
10 * version 2.1 of the License, or (at your option) any later version. | |
11 * | |
12 * FFmpeg is distributed in the hope that it will be useful, | |
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
18 * License along with FFmpeg; if not, write to the Free Software | |
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 */ | |
8150 | 21 |
8096 | 22 /** |
23 * @file qcelpdec.c | |
24 * QCELP decoder | |
25 * @author Reynaldo H. Verdejo Pinochet | |
8151
e2c068cb210a
Credit Kenan Gillet for his contributions towards merging the SoC QCELP decoder.
reynaldo
parents:
8150
diff
changeset
|
26 * @remark FFmpeg merging spearheaded by Kenan Gillet |
8096 | 27 */ |
28 | |
29 #include <stddef.h> | |
30 | |
31 #include "avcodec.h" | |
32 #include "bitstream.h" | |
33 | |
34 #include "qcelpdata.h" | |
35 | |
36 #include "celp_math.h" | |
37 #include "celp_filters.h" | |
38 | |
39 #undef NDEBUG | |
40 #include <assert.h> | |
41 | |
8238 | 42 typedef enum |
43 { | |
44 I_F_Q = -1, /*!< insufficient frame quality */ | |
45 SILENCE, | |
46 RATE_OCTAVE, | |
47 RATE_QUARTER, | |
48 RATE_HALF, | |
49 RATE_FULL | |
50 } qcelp_packet_rate; | |
51 | |
8230 | 52 typedef struct { |
53 GetBitContext gb; | |
54 qcelp_packet_rate bitrate; | |
55 QCELPFrame frame; /*!< unpacked data frame */ | |
56 uint8_t erasure_count; | |
57 uint8_t octave_count; /*!< count the consecutive RATE_OCTAVE frames */ | |
58 float prev_lspf[10]; | |
59 float predictor_lspf[10]; /*!< LSP predictor, | |
60 only use for RATE_OCTAVE and I_F_Q */ | |
8238 | 61 float pitch_synthesis_filter_mem[303]; |
62 float pitch_pre_filter_mem[303]; | |
63 float rnd_fir_filter_mem[180]; | |
8230 | 64 float formant_mem[170]; |
65 float last_codebook_gain; | |
66 int prev_g1[2]; | |
67 int prev_bitrate; | |
8240
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
68 float pitch_gain[4]; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
69 uint8_t pitch_lag[4]; |
8230 | 70 uint16_t first16bits; |
71 } QCELPContext; | |
72 | |
8238 | 73 /** |
74 * Reconstructs LPC coefficients from the line spectral pair frequencies. | |
75 * | |
76 * TIA/EIA/IS-733 2.4.3.3.5 | |
77 */ | |
8240
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
78 void ff_qcelp_lspf2lpc(const float *lspf, float *lpc); |
8238 | 79 |
8150 | 80 static void weighted_vector_sumf(float *out, const float *in_a, |
81 const float *in_b, float weight_coeff_a, | |
82 float weight_coeff_b, int length) | |
83 { | |
84 int i; | |
8123 | 85 |
8150 | 86 for(i=0; i<length; i++) |
8123 | 87 out[i] = weight_coeff_a * in_a[i] |
88 + weight_coeff_b * in_b[i]; | |
89 } | |
90 | |
8096 | 91 /** |
8145 | 92 * Initialize the speech codec according to the specification. |
93 * | |
94 * TIA/EIA/IS-733 2.4.9 | |
95 */ | |
8192 | 96 static av_cold int qcelp_decode_init(AVCodecContext *avctx) |
97 { | |
8145 | 98 QCELPContext *q = avctx->priv_data; |
99 int i; | |
100 | |
101 avctx->sample_fmt = SAMPLE_FMT_FLT; | |
102 | |
103 for (i = 0; i < 10; i++) | |
104 q->prev_lspf[i] = (i + 1) / 11.; | |
105 | |
106 return 0; | |
107 } | |
108 | |
109 /** | |
8191 | 110 * Decodes the 10 quantized LSP frequencies from the LSPV/LSP |
111 * transmission codes of any bitrate and checks for badly received packets. | |
112 * | |
113 * @param q the context | |
114 * @param lspf line spectral pair frequencies | |
115 * | |
116 * @return 0 on success, -1 if the packet is badly received | |
117 * | |
118 * TIA/EIA/IS-733 2.4.3.2.6.2-2, 2.4.8.7.3 | |
119 */ | |
8192 | 120 static int decode_lspf(QCELPContext *q, float *lspf) |
121 { | |
8191 | 122 int i; |
123 float tmp_lspf; | |
124 | |
8192 | 125 if(q->bitrate == RATE_OCTAVE || q->bitrate == I_F_Q) |
126 { | |
8191 | 127 float smooth; |
128 const float *predictors = (q->prev_bitrate != RATE_OCTAVE && | |
129 q->prev_bitrate != I_F_Q ? q->prev_lspf | |
130 : q->predictor_lspf); | |
131 | |
8192 | 132 if(q->bitrate == RATE_OCTAVE) |
133 { | |
8191 | 134 q->octave_count++; |
135 | |
8192 | 136 for(i=0; i<10; i++) |
137 { | |
8191 | 138 q->predictor_lspf[i] = |
8230 | 139 lspf[i] = (q->frame.lspv[i] ? QCELP_LSP_SPREAD_FACTOR |
140 : -QCELP_LSP_SPREAD_FACTOR) | |
8191 | 141 + predictors[i] * QCELP_LSP_OCTAVE_PREDICTOR |
142 + (i + 1) * ((1 - QCELP_LSP_OCTAVE_PREDICTOR)/11); | |
143 } | |
144 smooth = (q->octave_count < 10 ? .875 : 0.1); | |
8192 | 145 }else |
146 { | |
8191 | 147 float erasure_coeff = QCELP_LSP_OCTAVE_PREDICTOR; |
148 | |
149 assert(q->bitrate == I_F_Q); | |
150 | |
8192 | 151 if(q->erasure_count > 1) |
8191 | 152 erasure_coeff *= (q->erasure_count < 4 ? 0.9 : 0.7); |
153 | |
8192 | 154 for(i=0; i<10; i++) |
155 { | |
8191 | 156 q->predictor_lspf[i] = |
157 lspf[i] = (i + 1) * ( 1 - erasure_coeff)/11 | |
158 + erasure_coeff * predictors[i]; | |
159 } | |
160 smooth = 0.125; | |
161 } | |
162 | |
163 // Check the stability of the LSP frequencies. | |
164 lspf[0] = FFMAX(lspf[0], QCELP_LSP_SPREAD_FACTOR); | |
8192 | 165 for(i=1; i<10; i++) |
8191 | 166 lspf[i] = FFMAX(lspf[i], (lspf[i-1] + QCELP_LSP_SPREAD_FACTOR)); |
167 | |
168 lspf[9] = FFMIN(lspf[9], (1.0 - QCELP_LSP_SPREAD_FACTOR)); | |
8192 | 169 for(i=9; i>0; i--) |
8191 | 170 lspf[i-1] = FFMIN(lspf[i-1], (lspf[i] - QCELP_LSP_SPREAD_FACTOR)); |
171 | |
172 // Low-pass filter the LSP frequencies. | |
8192 | 173 weighted_vector_sumf(lspf, lspf, q->prev_lspf, smooth, 1.0-smooth, 10); |
174 }else | |
175 { | |
8191 | 176 q->octave_count = 0; |
177 | |
178 tmp_lspf = 0.; | |
8192 | 179 for(i=0; i<5 ; i++) |
180 { | |
8230 | 181 lspf[2*i+0] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][0] * 0.0001; |
182 lspf[2*i+1] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][1] * 0.0001; | |
8191 | 183 } |
184 | |
185 // Check for badly received packets. | |
8192 | 186 if(q->bitrate == RATE_QUARTER) |
187 { | |
188 if(lspf[9] <= .70 || lspf[9] >= .97) | |
8191 | 189 return -1; |
8192 | 190 for(i=3; i<10; i++) |
191 if(fabs(lspf[i] - lspf[i-2]) < .08) | |
8191 | 192 return -1; |
8192 | 193 }else |
194 { | |
195 if(lspf[9] <= .66 || lspf[9] >= .985) | |
8191 | 196 return -1; |
8192 | 197 for(i=4; i<10; i++) |
8191 | 198 if (fabs(lspf[i] - lspf[i-4]) < .0931) |
199 return -1; | |
200 } | |
201 } | |
202 return 0; | |
203 } | |
204 | |
205 /** | |
8230 | 206 * Converts codebook transmission codes to GAIN and INDEX. |
207 * | |
208 * @param q the context | |
209 * @param gain array holding the decoded gain | |
210 * | |
211 * TIA/EIA/IS-733 2.4.6.2 | |
212 */ | |
213 static void decode_gain_and_index(QCELPContext *q, | |
214 float *gain) { | |
215 int i, subframes_count, g1[16]; | |
216 float slope; | |
217 | |
218 if (q->bitrate >= RATE_QUARTER) { | |
219 switch (q->bitrate) { | |
220 case RATE_FULL: subframes_count = 16; break; | |
221 case RATE_HALF: subframes_count = 4; break; | |
222 default: subframes_count = 5; | |
223 } | |
224 for (i = 0; i < subframes_count; i++) { | |
225 g1[i] = 4 * q->frame.cbgain[i]; | |
226 if (q->bitrate == RATE_FULL && !((i+1) & 3)) { | |
227 g1[i] += av_clip((g1[i-1] + g1[i-2] + g1[i-3]) / 3 - 6, 0, 32); | |
228 } | |
229 | |
230 gain[i] = qcelp_g12ga[g1[i]]; | |
231 | |
232 if (q->frame.cbsign[i]) { | |
233 gain[i] = -gain[i]; | |
234 q->frame.cindex[i] = (q->frame.cindex[i]-89) & 127; | |
235 } | |
236 } | |
237 | |
238 q->prev_g1[0] = g1[i-2]; | |
239 q->prev_g1[1] = g1[i-1]; | |
240 q->last_codebook_gain = qcelp_g12ga[g1[i-1]]; | |
241 | |
242 if (q->bitrate == RATE_QUARTER) { | |
243 // Provide smoothing of the unvoiced excitation energy. | |
244 gain[7] = gain[4]; | |
245 gain[6] = 0.4*gain[3] + 0.6*gain[4]; | |
246 gain[5] = gain[3]; | |
247 gain[4] = 0.8*gain[2] + 0.2*gain[3]; | |
248 gain[3] = 0.2*gain[1] + 0.8*gain[2]; | |
249 gain[2] = gain[1]; | |
250 gain[1] = 0.6*gain[0] + 0.4*gain[1]; | |
251 } | |
252 } else { | |
253 if (q->bitrate == RATE_OCTAVE) { | |
254 g1[0] = 2 * q->frame.cbgain[0] | |
255 + av_clip((q->prev_g1[0] + q->prev_g1[1]) / 2 - 5, 0, 54); | |
256 subframes_count = 8; | |
257 } else { | |
258 assert(q->bitrate == I_F_Q); | |
259 | |
260 g1[0] = q->prev_g1[1]; | |
261 switch (q->erasure_count) { | |
262 case 1 : break; | |
263 case 2 : g1[0] -= 1; break; | |
264 case 3 : g1[0] -= 2; break; | |
265 default: g1[0] -= 6; | |
266 } | |
267 if (g1[0] < 0) | |
268 g1[0] = 0; | |
269 subframes_count = 4; | |
270 } | |
271 // This interpolation is done to produce smoother background noise. | |
272 slope = 0.5*(qcelp_g12ga[g1[0]] - q->last_codebook_gain) / subframes_count; | |
273 for (i = 1; i <= subframes_count; i++) | |
274 gain[i-1] = q->last_codebook_gain + slope * i; | |
275 q->last_codebook_gain = gain[i-2]; | |
276 | |
277 q->prev_g1[0] = q->prev_g1[1]; | |
278 q->prev_g1[1] = g1[0]; | |
279 } | |
280 } | |
281 | |
282 /** | |
8192 | 283 * If the received packet is Rate 1/4 a further sanity check is made of the |
284 * codebook gain. | |
8191 | 285 * |
286 * @param cbgain the unpacked cbgain array | |
287 * @return -1 if the sanity check fails, 0 otherwise | |
288 * | |
289 * TIA/EIA/IS-733 2.4.8.7.3 | |
290 */ | |
8192 | 291 static int codebook_sanity_check_for_rate_quarter(const uint8_t *cbgain) |
292 { | |
8193 | 293 int i, prev_diff=0; |
8191 | 294 |
8192 | 295 for(i=1; i<5; i++) |
296 { | |
297 int diff = cbgain[i] - cbgain[i-1]; | |
298 if(FFABS(diff) > 10) | |
299 return -1; | |
300 else if(FFABS(diff - prev_diff) > 12) | |
301 return -1; | |
302 prev_diff = diff; | |
8193 | 303 } |
304 return 0; | |
8191 | 305 } |
306 | |
307 /** | |
8145 | 308 * Computes the scaled codebook vector Cdn From INDEX and GAIN |
309 * for all rates. | |
310 * | |
311 * The specification lacks some information here. | |
312 * | |
313 * TIA/EIA/IS-733 has an omission on the codebook index determination | |
314 * formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says | |
315 * you have to subtract the decoded index parameter from the given scaled | |
316 * codebook vector index 'n' to get the desired circular codebook index, but | |
317 * it does not mention that you have to clamp 'n' to [0-9] in order to get | |
318 * RI-compliant results. | |
319 * | |
320 * The reason for this mistake seems to be the fact they forgot to mention you | |
321 * have to do these calculations per codebook subframe and adjust given | |
322 * equation values accordingly. | |
323 * | |
324 * @param q the context | |
325 * @param gain array holding the 4 pitch subframe gain values | |
326 * @param cdn_vector array for the generated scaled codebook vector | |
327 */ | |
8192 | 328 static void compute_svector(const QCELPContext *q, const float *gain, |
329 float *cdn_vector) | |
330 { | |
8145 | 331 int i, j, k; |
332 uint16_t cbseed, cindex; | |
333 float *rnd, tmp_gain, fir_filter_value; | |
334 | |
8192 | 335 switch(q->bitrate) |
336 { | |
337 case RATE_FULL: | |
338 for(i=0; i<16; i++) | |
339 { | |
340 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; | |
8230 | 341 cindex = -q->frame.cindex[i]; |
8192 | 342 for(j=0; j<10; j++) |
343 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127]; | |
344 } | |
8145 | 345 break; |
8192 | 346 case RATE_HALF: |
347 for(i=0; i<4; i++) | |
348 { | |
349 tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO; | |
8230 | 350 cindex = -q->frame.cindex[i]; |
8192 | 351 for (j = 0; j < 40; j++) |
8145 | 352 *cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127]; |
8192 | 353 } |
8145 | 354 break; |
8192 | 355 case RATE_QUARTER: |
8230 | 356 cbseed = (0x0003 & q->frame.lspv[4])<<14 | |
357 (0x003F & q->frame.lspv[3])<< 8 | | |
358 (0x0060 & q->frame.lspv[2])<< 1 | | |
359 (0x0007 & q->frame.lspv[1])<< 3 | | |
360 (0x0038 & q->frame.lspv[0])>> 3 ; | |
8192 | 361 rnd = q->rnd_fir_filter_mem + 20; |
362 for(i=0; i<8; i++) | |
363 { | |
364 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); | |
365 for(k=0; k<20; k++) | |
366 { | |
367 cbseed = 521 * cbseed + 259; | |
368 *rnd = (int16_t)cbseed; | |
8145 | 369 |
8192 | 370 // FIR filter |
371 fir_filter_value = 0.0; | |
372 for(j=0; j<10; j++) | |
373 fir_filter_value += qcelp_rnd_fir_coefs[j ] | |
374 * (rnd[-j ] + rnd[-20+j]); | |
8145 | 375 |
8192 | 376 fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10]; |
377 *cdn_vector++ = tmp_gain * fir_filter_value; | |
378 rnd++; | |
379 } | |
8145 | 380 } |
8192 | 381 memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float)); |
8145 | 382 break; |
8192 | 383 case RATE_OCTAVE: |
384 cbseed = q->first16bits; | |
385 for(i=0; i<8; i++) | |
386 { | |
387 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); | |
388 for(j=0; j<20; j++) | |
389 { | |
390 cbseed = 521 * cbseed + 259; | |
391 *cdn_vector++ = tmp_gain * (int16_t)cbseed; | |
392 } | |
8145 | 393 } |
394 break; | |
8192 | 395 case I_F_Q: |
396 cbseed = -44; // random codebook index | |
397 for(i=0; i<4; i++) | |
398 { | |
399 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; | |
400 for(j=0; j<40; j++) | |
401 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127]; | |
402 } | |
8145 | 403 break; |
404 } | |
405 } | |
406 | |
407 /** | |
408 * Apply generic gain control. | |
409 * | |
410 * @param v_out output vector | |
411 * @param v_in gain-controlled vector | |
412 * @param v_ref vector to control gain of | |
413 * | |
414 * FIXME: If v_ref is a zero vector, it energy is zero | |
415 * and the behavior of the gain control is | |
416 * undefined in the specs. | |
417 * | |
418 * TIA/EIA/IS-733 2.4.8.3-2/3/4/5, 2.4.8.6 | |
419 */ | |
8192 | 420 static void apply_gain_ctrl(float *v_out, const float *v_ref, |
421 const float *v_in) | |
422 { | |
8145 | 423 int i, j, len; |
424 float scalefactor; | |
425 | |
8192 | 426 for(i=0, j=0; i<4; i++) |
427 { | |
8145 | 428 scalefactor = ff_dot_productf(v_in + j, v_in + j, 40); |
8192 | 429 if(scalefactor) |
430 scalefactor = sqrt(ff_dot_productf(v_ref + j, v_ref + j, 40) | |
431 / scalefactor); | |
8145 | 432 else |
433 av_log_missing_feature(NULL, "Zero energy for gain control", 1); | |
8192 | 434 for(len=j+40; j<len; j++) |
8145 | 435 v_out[j] = scalefactor * v_in[j]; |
436 } | |
437 } | |
438 | |
439 /** | |
8096 | 440 * Apply filter in pitch-subframe steps. |
441 * | |
442 * @param memory buffer for the previous state of the filter | |
443 * - must be able to contain 303 elements | |
444 * - the 143 first elements are from the previous state | |
445 * - the next 160 are for output | |
446 * @param v_in input filter vector | |
447 * @param gain per-subframe gain array, each element is between 0.0 and 2.0 | |
448 * @param lag per-subframe lag array, each element is | |
449 * - between 16 and 143 if its corresponding pfrac is 0, | |
450 * - between 16 and 139 otherwise | |
8192 | 451 * @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0 |
452 * otherwise | |
8096 | 453 * |
454 * @return filter output vector | |
455 */ | |
8150 | 456 static const float *do_pitchfilter(float memory[303], const float v_in[160], |
457 const float gain[4], const uint8_t *lag, | |
458 const uint8_t pfrac[4]) | |
459 { | |
8096 | 460 int i, j; |
461 float *v_lag, *v_out; | |
462 const float *v_len; | |
463 | |
464 v_out = memory + 143; // Output vector starts at memory[143]. | |
465 | |
8150 | 466 for(i=0; i<4; i++) |
467 { | |
468 if(gain[i]) | |
469 { | |
8096 | 470 v_lag = memory + 143 + 40 * i - lag[i]; |
8150 | 471 for(v_len=v_in+40; v_in<v_len; v_in++) |
472 { | |
473 if(pfrac[i]) // If it is a fractional lag... | |
474 { | |
475 for(j=0, *v_out=0.; j<4; j++) | |
8096 | 476 *v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]); |
8150 | 477 }else |
8096 | 478 *v_out = *v_lag; |
479 | |
480 *v_out = *v_in + gain[i] * *v_out; | |
481 | |
482 v_lag++; | |
483 v_out++; | |
484 } | |
8150 | 485 }else |
486 { | |
8096 | 487 memcpy(v_out, v_in, 40 * sizeof(float)); |
488 v_in += 40; | |
489 v_out += 40; | |
490 } | |
8150 | 491 } |
8096 | 492 |
493 memmove(memory, memory + 160, 143 * sizeof(float)); | |
494 return memory + 143; | |
495 } | |
496 | |
8127 | 497 /** |
8240
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
498 * Apply pitch synthesis filter and pitch prefilter to the scaled codebook vector. |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
499 * TIA/EIA/IS-733 2.4.5.2 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
500 * |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
501 * @param q the context |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
502 * @param cdn_vector the scaled codebook vector |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
503 */ |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
504 static void apply_pitch_filters(QCELPContext *q, |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
505 float *cdn_vector) { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
506 int i; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
507 const float *v_synthesis_filtered, *v_pre_filtered; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
508 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
509 if (q->bitrate >= RATE_HALF || |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
510 (q->bitrate == I_F_Q && (q->prev_bitrate >= RATE_HALF))) { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
511 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
512 if (q->bitrate >= RATE_HALF) { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
513 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
514 // Compute gain & lag for the whole frame. |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
515 for (i = 0; i < 4; i++) { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
516 q->pitch_gain[i] = q->frame.plag[i] ? (q->frame.pgain[i] + 1) * 0.25 : 0.0; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
517 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
518 q->pitch_lag[i] = q->frame.plag[i] + 16; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
519 } |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
520 } else { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
521 float max_pitch_gain = q->erasure_count < 3 ? 0.9 - 0.3 * (q->erasure_count - 1) |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
522 : 0.0; |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
523 for (i = 0; i < 4; i++) |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
524 q->pitch_gain[i] = FFMIN(q->pitch_gain[i], max_pitch_gain); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
525 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
526 memset(q->frame.pfrac, 0, sizeof(q->frame.pfrac)); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
527 } |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
528 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
529 // pitch synthesis filter |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
530 v_synthesis_filtered = do_pitchfilter(q->pitch_synthesis_filter_mem, cdn_vector, |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
531 q->pitch_gain, q->pitch_lag, q->frame.pfrac); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
532 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
533 // pitch prefilter update |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
534 for (i = 0; i < 4; i++) |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
535 q->pitch_gain[i] = 0.5 * FFMIN(q->pitch_gain[i], 1.0); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
536 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
537 v_pre_filtered = do_pitchfilter(q->pitch_pre_filter_mem, v_synthesis_filtered, |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
538 q->pitch_gain, q->pitch_lag, q->frame.pfrac); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
539 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
540 apply_gain_ctrl(cdn_vector, v_synthesis_filtered, v_pre_filtered); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
541 } else { |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
542 memcpy(q->pitch_synthesis_filter_mem, cdn_vector + 17, 143 * sizeof(float)); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
543 memcpy(q->pitch_pre_filter_mem, cdn_vector + 17, 143 * sizeof(float)); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
544 memset(q->pitch_gain, 0, sizeof(q->pitch_gain)); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
545 memset(q->pitch_lag, 0, sizeof(q->pitch_lag)); |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
546 } |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
547 } |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
548 |
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
549 /** |
8127 | 550 * Interpolates LSP frequencies and computes LPC coefficients |
8191 | 551 * for a given bitrate & pitch subframe. |
8127 | 552 * |
553 * TIA/EIA/IS-733 2.4.3.3.4 | |
554 * | |
555 * @param q the context | |
556 * @param curr_lspf LSP frequencies vector of the current frame | |
557 * @param lpc float vector for the resulting LPC | |
558 * @param subframe_num frame number in decoded stream | |
559 */ | |
8150 | 560 void interpolate_lpc(QCELPContext *q, const float *curr_lspf, float *lpc, |
561 const int subframe_num) | |
562 { | |
8127 | 563 float interpolated_lspf[10]; |
564 float weight; | |
565 | |
8191 | 566 if(q->bitrate >= RATE_QUARTER) |
8127 | 567 weight = 0.25 * (subframe_num + 1); |
8191 | 568 else if(q->bitrate == RATE_OCTAVE && !subframe_num) |
8127 | 569 weight = 0.625; |
8150 | 570 else |
8127 | 571 weight = 1.0; |
572 | |
8150 | 573 if(weight != 1.0) |
574 { | |
575 weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf, | |
576 weight, 1.0 - weight, 10); | |
8240
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
577 ff_qcelp_lspf2lpc(interpolated_lspf, lpc); |
8191 | 578 }else if(q->bitrate >= RATE_QUARTER || (q->bitrate == I_F_Q && !subframe_num)) |
8240
d3d0d9cc0e50
Commit last ok'ed parts of QCELP decoder and enable it.
vitor
parents:
8238
diff
changeset
|
579 ff_qcelp_lspf2lpc(curr_lspf, lpc); |
8127 | 580 } |
581 | |
8191 | 582 static int buf_size2bitrate(const int buf_size) |
8150 | 583 { |
584 switch(buf_size) | |
585 { | |
586 case 35: | |
587 return RATE_FULL; | |
588 case 17: | |
589 return RATE_HALF; | |
590 case 8: | |
591 return RATE_QUARTER; | |
592 case 4: | |
593 return RATE_OCTAVE; | |
594 case 1: | |
595 return SILENCE; | |
8123 | 596 } |
8150 | 597 |
8123 | 598 return -1; |
599 } | |
600 | |
8238 | 601 /** |
602 * Determine the bitrate from the frame size and/or the first byte of the frame. | |
603 * | |
604 * @param avctx the AV codec context | |
605 * @param buf_size length of the buffer | |
606 * @param buf the bufffer | |
607 * | |
608 * @return the bitrate on success, | |
609 * I_F_Q if the bitrate cannot be satisfactorily determined | |
610 * | |
611 * TIA/EIA/IS-733 2.4.8.7.1 | |
612 */ | |
613 static int determine_bitrate(AVCodecContext *avctx, | |
614 const int buf_size, | |
615 uint8_t **buf) { | |
616 qcelp_packet_rate bitrate; | |
617 | |
618 if ((bitrate = buf_size2bitrate(buf_size)) >= 0) { | |
619 if (bitrate > **buf) { | |
620 av_log(avctx, AV_LOG_WARNING, "Claimed bitrate and buffer size mismatch.\n"); | |
621 bitrate = **buf; | |
622 } else if (bitrate < **buf) { | |
623 av_log(avctx, AV_LOG_ERROR, "Buffer is too small for the claimed bitrate.\n"); | |
624 return I_F_Q; | |
625 } | |
626 (*buf)++; | |
627 } else if ((bitrate = buf_size2bitrate(buf_size + 1)) >= 0) { | |
628 av_log(avctx, AV_LOG_WARNING, | |
629 "Bitrate byte is missing, guessing the bitrate from packet size.\n"); | |
630 } else | |
631 return I_F_Q; | |
632 | |
633 if (bitrate == SILENCE) { | |
634 // FIXME: the decoder should not handle SILENCE frames as I_F_Q frames | |
635 av_log_missing_feature(avctx, "Blank frame", 1); | |
636 bitrate = I_F_Q; | |
637 } | |
638 return bitrate; | |
639 } | |
640 | |
8096 | 641 static void warn_insufficient_frame_quality(AVCodecContext *avctx, |
8150 | 642 const char *message) |
643 { | |
644 av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number, | |
645 message); | |
8096 | 646 } |
8127 | 647 |
8230 | 648 static int qcelp_decode_frame(AVCodecContext *avctx, |
649 void *data, | |
650 int *data_size, | |
651 uint8_t *buf, | |
652 const int buf_size) { | |
653 QCELPContext *q = avctx->priv_data; | |
654 float *outbuffer = data; | |
655 int i; | |
656 float quantized_lspf[10], lpc[10]; | |
657 float gain[16]; | |
658 float *formant_mem; | |
659 | |
660 if ((q->bitrate = determine_bitrate(avctx, buf_size, &buf)) == I_F_Q) { | |
661 warn_insufficient_frame_quality(avctx, "bitrate cannot be determined."); | |
662 goto erasure; | |
663 } | |
664 | |
665 if (q->bitrate == RATE_OCTAVE && | |
666 (q->first16bits = AV_RB16(buf)) == 0xFFFF) { | |
667 warn_insufficient_frame_quality(avctx, "Bitrate is 1/8 and first 16 bits are on."); | |
668 goto erasure; | |
669 } | |
670 | |
671 if (q->bitrate > SILENCE) { | |
672 const QCELPBitmap *bitmaps = qcelp_unpacking_bitmaps_per_rate[q->bitrate]; | |
673 const QCELPBitmap *bitmaps_end = qcelp_unpacking_bitmaps_per_rate[q->bitrate] | |
674 + qcelp_unpacking_bitmaps_lengths[q->bitrate]; | |
675 uint8_t *unpacked_data = (uint8_t *)&q->frame; | |
676 | |
677 init_get_bits(&q->gb, buf, 8*buf_size); | |
678 | |
679 memset(&q->frame, 0, sizeof(QCELPFrame)); | |
680 | |
681 for (; bitmaps < bitmaps_end; bitmaps++) | |
682 unpacked_data[bitmaps->index] |= get_bits(&q->gb, bitmaps->bitlen) << bitmaps->bitpos; | |
683 | |
684 // Check for erasures/blanks on rates 1, 1/4 and 1/8. | |
685 if (q->frame.reserved) { | |
686 warn_insufficient_frame_quality(avctx, "Wrong data in reserved frame area."); | |
687 goto erasure; | |
688 } | |
689 if (q->bitrate == RATE_QUARTER && codebook_sanity_check_for_rate_quarter(q->frame.cbgain)) { | |
690 warn_insufficient_frame_quality(avctx, "Codebook gain sanity check failed."); | |
691 goto erasure; | |
692 } | |
693 | |
694 if (q->bitrate >= RATE_HALF) { | |
695 for (i = 0; i < 4; i++) { | |
696 if (q->frame.pfrac[i] && q->frame.plag[i] >= 124) { | |
697 warn_insufficient_frame_quality(avctx, "Cannot initialize pitch filter."); | |
698 goto erasure; | |
699 } | |
700 } | |
701 } | |
702 } | |
703 | |
704 decode_gain_and_index(q, gain); | |
705 compute_svector(q, gain, outbuffer); | |
706 | |
707 if (decode_lspf(q, quantized_lspf) < 0) { | |
708 warn_insufficient_frame_quality(avctx, "Badly received packets in frame."); | |
709 goto erasure; | |
710 } | |
711 | |
712 | |
713 apply_pitch_filters(q, outbuffer); | |
714 | |
715 if (q->bitrate == I_F_Q) { | |
716 erasure: | |
717 q->bitrate = I_F_Q; | |
718 q->erasure_count++; | |
719 decode_gain_and_index(q, gain); | |
720 compute_svector(q, gain, outbuffer); | |
721 decode_lspf(q, quantized_lspf); | |
722 apply_pitch_filters(q, outbuffer); | |
723 } else | |
724 q->erasure_count = 0; | |
725 | |
726 formant_mem = q->formant_mem + 10; | |
727 for (i = 0; i < 4; i++) { | |
728 interpolate_lpc(q, quantized_lspf, lpc, i); | |
729 ff_celp_lp_synthesis_filterf(formant_mem, lpc, outbuffer + i * 40, 40, 10); | |
730 formant_mem += 40; | |
731 } | |
732 memcpy(q->formant_mem, q->formant_mem + 160, 10 * sizeof(float)); | |
733 | |
734 // FIXME: postfilter and final gain control should be here. | |
735 // TIA/EIA/IS-733 2.4.8.6 | |
736 | |
737 formant_mem = q->formant_mem + 10; | |
738 for (i = 0; i < 160; i++) | |
739 *outbuffer++ = av_clipf(*formant_mem++, QCELP_CLIP_LOWER_BOUND, QCELP_CLIP_UPPER_BOUND); | |
740 | |
741 memcpy(q->prev_lspf, quantized_lspf, sizeof(q->prev_lspf)); | |
742 q->prev_bitrate = q->bitrate; | |
743 | |
744 *data_size = 160 * sizeof(*outbuffer); | |
745 | |
746 return *data_size; | |
747 } | |
748 | |
8127 | 749 AVCodec qcelp_decoder = |
750 { | |
751 .name = "qcelp", | |
752 .type = CODEC_TYPE_AUDIO, | |
753 .id = CODEC_ID_QCELP, | |
754 .init = qcelp_decode_init, | |
755 .decode = qcelp_decode_frame, | |
756 .priv_data_size = sizeof(QCELPContext), | |
757 .long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"), | |
758 }; |