FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/evrcdec.c
Date: 2021-09-23 20:34:37
Exec Total Coverage
Lines: 0 433 0.0%
Branches: 0 252 0.0%

Line Branch Exec Source
1 /*
2 * Enhanced Variable Rate Codec, Service Option 3 decoder
3 * Copyright (c) 2013 Paul B Mahol
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 */
21
22 /**
23 * @file
24 * Enhanced Variable Rate Codec, Service Option 3 decoder
25 * @author Paul B Mahol
26 */
27
28 #include "libavutil/channel_layout.h"
29 #include "libavutil/mathematics.h"
30 #include "libavutil/opt.h"
31 #include "avcodec.h"
32 #include "internal.h"
33 #include "get_bits.h"
34 #include "evrcdata.h"
35 #include "acelp_vectors.h"
36 #include "lsp.h"
37
38 #define MIN_LSP_SEP (0.05 / (2.0 * M_PI))
39 #define MIN_DELAY 20
40 #define MAX_DELAY 120
41 #define NB_SUBFRAMES 3
42 #define SUBFRAME_SIZE 54
43 #define FILTER_ORDER 10
44 #define ACB_SIZE 128
45
46 typedef enum {
47 RATE_ERRS = -1,
48 SILENCE,
49 RATE_QUANT,
50 RATE_QUARTER,
51 RATE_HALF,
52 RATE_FULL,
53 } evrc_packet_rate;
54
55 /**
56 * EVRC-A unpacked data frame
57 */
58 typedef struct EVRCAFrame {
59 uint8_t lpc_flag; ///< spectral change indicator
60 uint16_t lsp[4]; ///< index into LSP codebook
61 uint8_t pitch_delay; ///< pitch delay for entire frame
62 uint8_t delay_diff; ///< delay difference for entire frame
63 uint8_t acb_gain[3]; ///< adaptive codebook gain
64 uint16_t fcb_shape[3][4]; ///< fixed codebook shape
65 uint8_t fcb_gain[3]; ///< fixed codebook gain index
66 uint8_t energy_gain; ///< frame energy gain index
67 uint8_t tty; ///< tty baud rate bit
68 } EVRCAFrame;
69
70 typedef struct EVRCContext {
71 AVClass *class;
72
73 int postfilter;
74
75 GetBitContext gb;
76 evrc_packet_rate bitrate;
77 evrc_packet_rate last_valid_bitrate;
78 EVRCAFrame frame;
79
80 float lspf[FILTER_ORDER];
81 float prev_lspf[FILTER_ORDER];
82 float synthesis[FILTER_ORDER];
83 float postfilter_fir[FILTER_ORDER];
84 float postfilter_iir[FILTER_ORDER];
85 float postfilter_residual[ACB_SIZE + SUBFRAME_SIZE];
86 float pitch_delay;
87 float prev_pitch_delay;
88 float avg_acb_gain; ///< average adaptive codebook gain
89 float avg_fcb_gain; ///< average fixed codebook gain
90 float pitch[ACB_SIZE + FILTER_ORDER + SUBFRAME_SIZE];
91 float pitch_back[ACB_SIZE];
92 float interpolation_coeffs[136];
93 float energy_vector[NB_SUBFRAMES];
94 float fade_scale;
95 float last;
96
97 uint8_t prev_energy_gain;
98 uint8_t prev_error_flag;
99 uint8_t warned_buf_mismatch_bitrate;
100 } EVRCContext;
101
102 /**
103 * Frame unpacking for RATE_FULL, RATE_HALF and RATE_QUANT
104 *
105 * @param e the context
106 *
107 * TIA/IS-127 Table 4.21-1
108 */
109 static void unpack_frame(EVRCContext *e)
110 {
111 EVRCAFrame *frame = &e->frame;
112 GetBitContext *gb = &e->gb;
113
114 switch (e->bitrate) {
115 case RATE_FULL:
116 frame->lpc_flag = get_bits1(gb);
117 frame->lsp[0] = get_bits(gb, 6);
118 frame->lsp[1] = get_bits(gb, 6);
119 frame->lsp[2] = get_bits(gb, 9);
120 frame->lsp[3] = get_bits(gb, 7);
121 frame->pitch_delay = get_bits(gb, 7);
122 frame->delay_diff = get_bits(gb, 5);
123 frame->acb_gain[0] = get_bits(gb, 3);
124 frame->fcb_shape[0][0] = get_bits(gb, 8);
125 frame->fcb_shape[0][1] = get_bits(gb, 8);
126 frame->fcb_shape[0][2] = get_bits(gb, 8);
127 frame->fcb_shape[0][3] = get_bits(gb, 11);
128 frame->fcb_gain[0] = get_bits(gb, 5);
129 frame->acb_gain[1] = get_bits(gb, 3);
130 frame->fcb_shape[1][0] = get_bits(gb, 8);
131 frame->fcb_shape[1][1] = get_bits(gb, 8);
132 frame->fcb_shape[1][2] = get_bits(gb, 8);
133 frame->fcb_shape[1][3] = get_bits(gb, 11);
134 frame->fcb_gain [1] = get_bits(gb, 5);
135 frame->acb_gain [2] = get_bits(gb, 3);
136 frame->fcb_shape[2][0] = get_bits(gb, 8);
137 frame->fcb_shape[2][1] = get_bits(gb, 8);
138 frame->fcb_shape[2][2] = get_bits(gb, 8);
139 frame->fcb_shape[2][3] = get_bits(gb, 11);
140 frame->fcb_gain [2] = get_bits(gb, 5);
141 frame->tty = get_bits1(gb);
142 break;
143 case RATE_HALF:
144 frame->lsp [0] = get_bits(gb, 7);
145 frame->lsp [1] = get_bits(gb, 7);
146 frame->lsp [2] = get_bits(gb, 8);
147 frame->pitch_delay = get_bits(gb, 7);
148 frame->acb_gain [0] = get_bits(gb, 3);
149 frame->fcb_shape[0][0] = get_bits(gb, 10);
150 frame->fcb_gain [0] = get_bits(gb, 4);
151 frame->acb_gain [1] = get_bits(gb, 3);
152 frame->fcb_shape[1][0] = get_bits(gb, 10);
153 frame->fcb_gain [1] = get_bits(gb, 4);
154 frame->acb_gain [2] = get_bits(gb, 3);
155 frame->fcb_shape[2][0] = get_bits(gb, 10);
156 frame->fcb_gain [2] = get_bits(gb, 4);
157 break;
158 case RATE_QUANT:
159 frame->lsp [0] = get_bits(gb, 4);
160 frame->lsp [1] = get_bits(gb, 4);
161 frame->energy_gain = get_bits(gb, 8);
162 break;
163 }
164 }
165
166 static evrc_packet_rate buf_size2bitrate(const int buf_size)
167 {
168 switch (buf_size) {
169 case 23: return RATE_FULL;
170 case 11: return RATE_HALF;
171 case 6: return RATE_QUARTER;
172 case 3: return RATE_QUANT;
173 case 1: return SILENCE;
174 }
175
176 return RATE_ERRS;
177 }
178
179 /**
180 * Determine the bitrate from the frame size and/or the first byte of the frame.
181 *
182 * @param avctx the AV codec context
183 * @param buf_size length of the buffer
184 * @param buf the bufffer
185 *
186 * @return the bitrate on success,
187 * RATE_ERRS if the bitrate cannot be satisfactorily determined
188 */
189 static evrc_packet_rate determine_bitrate(AVCodecContext *avctx,
190 int *buf_size,
191 const uint8_t **buf)
192 {
193 evrc_packet_rate bitrate;
194
195 if ((bitrate = buf_size2bitrate(*buf_size)) >= 0) {
196 if (bitrate > **buf) {
197 EVRCContext *e = avctx->priv_data;
198 if (!e->warned_buf_mismatch_bitrate) {
199 av_log(avctx, AV_LOG_WARNING,
200 "Claimed bitrate and buffer size mismatch.\n");
201 e->warned_buf_mismatch_bitrate = 1;
202 }
203 bitrate = **buf;
204 } else if (bitrate < **buf) {
205 av_log(avctx, AV_LOG_ERROR,
206 "Buffer is too small for the claimed bitrate.\n");
207 return RATE_ERRS;
208 }
209 (*buf)++;
210 *buf_size -= 1;
211 } else if ((bitrate = buf_size2bitrate(*buf_size + 1)) >= 0) {
212 av_log(avctx, AV_LOG_DEBUG,
213 "Bitrate byte is missing, guessing the bitrate from packet size.\n");
214 } else
215 return RATE_ERRS;
216
217 return bitrate;
218 }
219
220 static void warn_insufficient_frame_quality(AVCodecContext *avctx,
221 const char *message)
222 {
223 av_log(avctx, AV_LOG_WARNING, "Frame #%d, %s\n",
224 avctx->frame_number, message);
225 }
226
227 /**
228 * Initialize the speech codec according to the specification.
229 *
230 * TIA/IS-127 5.2
231 */
232 static av_cold int evrc_decode_init(AVCodecContext *avctx)
233 {
234 EVRCContext *e = avctx->priv_data;
235 int i, n, idx = 0;
236 float denom = 2.0 / (2.0 * 8.0 + 1.0);
237
238 avctx->channels = 1;
239 avctx->channel_layout = AV_CH_LAYOUT_MONO;
240 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
241
242 for (i = 0; i < FILTER_ORDER; i++) {
243 e->prev_lspf[i] = (i + 1) * 0.048;
244 e->synthesis[i] = 0.0;
245 }
246
247 for (i = 0; i < ACB_SIZE; i++)
248 e->pitch[i] = e->pitch_back[i] = 0.0;
249
250 e->last_valid_bitrate = RATE_QUANT;
251 e->prev_pitch_delay = 40.0;
252 e->fade_scale = 1.0;
253 e->prev_error_flag = 0;
254 e->avg_acb_gain = e->avg_fcb_gain = 0.0;
255
256 for (i = 0; i < 8; i++) {
257 float tt = ((float)i - 8.0 / 2.0) / 8.0;
258
259 for (n = -8; n <= 8; n++, idx++) {
260 float arg1 = M_PI * 0.9 * (tt - n);
261 float arg2 = M_PI * (tt - n);
262
263 e->interpolation_coeffs[idx] = 0.9;
264 if (arg1)
265 e->interpolation_coeffs[idx] *= (0.54 + 0.46 * cos(arg2 * denom)) *
266 sin(arg1) / arg1;
267 }
268 }
269
270 return 0;
271 }
272
273 /**
274 * Decode the 10 vector quantized line spectral pair frequencies from the LSP
275 * transmission codes of any bitrate and check for badly received packets.
276 *
277 * @param e the context
278 *
279 * @return 0 on success, -1 if the packet is badly received
280 *
281 * TIA/IS-127 5.2.1, 5.7.1
282 */
283 static int decode_lspf(EVRCContext *e)
284 {
285 const float * const *codebooks = evrc_lspq_codebooks[e->bitrate];
286 int i, j, k = 0;
287
288 for (i = 0; i < evrc_lspq_nb_codebooks[e->bitrate]; i++) {
289 int row_size = evrc_lspq_codebooks_row_sizes[e->bitrate][i];
290 const float *codebook = codebooks[i];
291
292 for (j = 0; j < row_size; j++)
293 e->lspf[k++] = codebook[e->frame.lsp[i] * row_size + j];
294 }
295
296 // check for monotonic LSPs
297 for (i = 1; i < FILTER_ORDER; i++)
298 if (e->lspf[i] <= e->lspf[i - 1])
299 return -1;
300
301 // check for minimum separation of LSPs at the splits
302 for (i = 0, k = 0; i < evrc_lspq_nb_codebooks[e->bitrate] - 1; i++) {
303 k += evrc_lspq_codebooks_row_sizes[e->bitrate][i];
304 if (e->lspf[k] - e->lspf[k - 1] <= MIN_LSP_SEP)
305 return -1;
306 }
307
308 return 0;
309 }
310
311 /*
312 * Interpolation of LSP parameters.
313 *
314 * TIA/IS-127 5.2.3.1, 5.7.3.2
315 */
316 static void interpolate_lsp(float *ilsp, const float *lsp,
317 const float *prev, int index)
318 {
319 static const float lsp_interpolation_factors[] = { 0.1667, 0.5, 0.8333 };
320 ff_weighted_vector_sumf(ilsp, prev, lsp,
321 1.0 - lsp_interpolation_factors[index],
322 lsp_interpolation_factors[index], FILTER_ORDER);
323 }
324
325 /*
326 * Reconstruction of the delay contour.
327 *
328 * TIA/IS-127 5.2.2.3.2
329 */
330 static void interpolate_delay(float *dst, float current, float prev, int index)
331 {
332 static const float d_interpolation_factors[] = { 0, 0.3313, 0.6625, 1, 1 };
333 dst[0] = (1.0 - d_interpolation_factors[index ]) * prev
334 + d_interpolation_factors[index ] * current;
335 dst[1] = (1.0 - d_interpolation_factors[index + 1]) * prev
336 + d_interpolation_factors[index + 1] * current;
337 dst[2] = (1.0 - d_interpolation_factors[index + 2]) * prev
338 + d_interpolation_factors[index + 2] * current;
339 }
340
341 /*
342 * Convert the quantized, interpolated line spectral frequencies,
343 * to prediction coefficients.
344 *
345 * TIA/IS-127 5.2.3.2, 4.7.2.2
346 */
347 static void decode_predictor_coeffs(const float *ilspf, float *ilpc)
348 {
349 double lsp[FILTER_ORDER];
350 float a[FILTER_ORDER / 2 + 1], b[FILTER_ORDER / 2 + 1];
351 float a1[FILTER_ORDER / 2] = { 0 };
352 float a2[FILTER_ORDER / 2] = { 0 };
353 float b1[FILTER_ORDER / 2] = { 0 };
354 float b2[FILTER_ORDER / 2] = { 0 };
355 int i, k;
356
357 ff_acelp_lsf2lspd(lsp, ilspf, FILTER_ORDER);
358
359 for (k = 0; k <= FILTER_ORDER; k++) {
360 a[0] = k < 2 ? 0.25 : 0;
361 b[0] = k < 2 ? k < 1 ? 0.25 : -0.25 : 0;
362
363 for (i = 0; i < FILTER_ORDER / 2; i++) {
364 a[i + 1] = a[i] - 2 * lsp[i * 2 ] * a1[i] + a2[i];
365 b[i + 1] = b[i] - 2 * lsp[i * 2 + 1] * b1[i] + b2[i];
366 a2[i] = a1[i];
367 a1[i] = a[i];
368 b2[i] = b1[i];
369 b1[i] = b[i];
370 }
371
372 if (k)
373 ilpc[k - 1] = 2.0 * (a[FILTER_ORDER / 2] + b[FILTER_ORDER / 2]);
374 }
375 }
376
377 static void bl_intrp(EVRCContext *e, float *ex, float delay)
378 {
379 float *f;
380 int offset, i, coef_idx;
381 int16_t t;
382
383 offset = lrintf(delay);
384
385 t = (offset - delay + 0.5) * 8.0 + 0.5;
386 if (t == 8) {
387 t = 0;
388 offset--;
389 }
390
391 f = ex - offset - 8;
392
393 coef_idx = t * (2 * 8 + 1);
394
395 ex[0] = 0.0;
396 for (i = 0; i < 2 * 8 + 1; i++)
397 ex[0] += e->interpolation_coeffs[coef_idx + i] * f[i];
398 }
399
400 /*
401 * Adaptive codebook excitation.
402 *
403 * TIA/IS-127 5.2.2.3.3, 4.12.5.2
404 */
405 static void acb_excitation(EVRCContext *e, float *excitation, float gain,
406 const float delay[3], int length)
407 {
408 float denom, locdelay, dpr, invl;
409 int i;
410
411 invl = 1.0 / ((float) length);
412 dpr = length;
413
414 /* first at-most extra samples */
415 denom = (delay[1] - delay[0]) * invl;
416 for (i = 0; i < dpr; i++) {
417 locdelay = delay[0] + i * denom;
418 bl_intrp(e, excitation + i, locdelay);
419 }
420
421 denom = (delay[2] - delay[1]) * invl;
422 /* interpolation */
423 for (i = dpr; i < dpr + 10; i++) {
424 locdelay = delay[1] + (i - dpr) * denom;
425 bl_intrp(e, excitation + i, locdelay);
426 }
427
428 for (i = 0; i < length; i++)
429 excitation[i] *= gain;
430 }
431
432 static void decode_8_pulses_35bits(const uint16_t *fixed_index, float *cod)
433 {
434 int i, pos1, pos2, offset;
435
436 offset = (fixed_index[3] >> 9) & 3;
437
438 for (i = 0; i < 3; i++) {
439 pos1 = ((fixed_index[i] & 0x7f) / 11) * 5 + ((i + offset) % 5);
440 pos2 = ((fixed_index[i] & 0x7f) % 11) * 5 + ((i + offset) % 5);
441
442 cod[pos1] = (fixed_index[i] & 0x80) ? -1.0 : 1.0;
443
444 if (pos2 < pos1)
445 cod[pos2] = -cod[pos1];
446 else
447 cod[pos2] += cod[pos1];
448 }
449
450 pos1 = ((fixed_index[3] & 0x7f) / 11) * 5 + ((3 + offset) % 5);
451 pos2 = ((fixed_index[3] & 0x7f) % 11) * 5 + ((4 + offset) % 5);
452
453 cod[pos1] = (fixed_index[3] & 0x100) ? -1.0 : 1.0;
454 cod[pos2] = (fixed_index[3] & 0x80 ) ? -1.0 : 1.0;
455 }
456
457 static void decode_3_pulses_10bits(uint16_t fixed_index, float *cod)
458 {
459 float sign;
460 int pos;
461
462 sign = (fixed_index & 0x200) ? -1.0 : 1.0;
463
464 pos = ((fixed_index & 0x7) * 7) + 4;
465 cod[pos] += sign;
466 pos = (((fixed_index >> 3) & 0x7) * 7) + 2;
467 cod[pos] -= sign;
468 pos = (((fixed_index >> 6) & 0x7) * 7);
469 cod[pos] += sign;
470 }
471
472 /*
473 * Reconstruction of ACELP fixed codebook excitation for full and half rate.
474 *
475 * TIA/IS-127 5.2.3.7
476 */
477 static void fcb_excitation(EVRCContext *e, const uint16_t *codebook,
478 float *excitation, float pitch_gain,
479 int pitch_lag, int subframe_size)
480 {
481 int i;
482
483 if (e->bitrate == RATE_FULL)
484 decode_8_pulses_35bits(codebook, excitation);
485 else
486 decode_3_pulses_10bits(*codebook, excitation);
487
488 pitch_gain = av_clipf(pitch_gain, 0.2, 0.9);
489
490 for (i = pitch_lag; i < subframe_size; i++)
491 excitation[i] += pitch_gain * excitation[i - pitch_lag];
492 }
493
494 /**
495 * Synthesis of the decoder output signal.
496 *
497 * param[in] in input signal
498 * param[in] filter_coeffs LPC coefficients
499 * param[in/out] memory synthesis filter memory
500 * param buffer_length amount of data to process
501 * param[out] samples output samples
502 *
503 * TIA/IS-127 5.2.3.15, 5.7.3.4
504 */
505 static void synthesis_filter(const float *in, const float *filter_coeffs,
506 float *memory, int buffer_length, float *samples)
507 {
508 int i, j;
509
510 for (i = 0; i < buffer_length; i++) {
511 samples[i] = in[i];
512 for (j = FILTER_ORDER - 1; j > 0; j--) {
513 samples[i] -= filter_coeffs[j] * memory[j];
514 memory[j] = memory[j - 1];
515 }
516 samples[i] -= filter_coeffs[0] * memory[0];
517 memory[0] = samples[i];
518 }
519 }
520
521 static void bandwidth_expansion(float *coeff, const float *inbuf, float gamma)
522 {
523 double fac = gamma;
524 int i;
525
526 for (i = 0; i < FILTER_ORDER; i++) {
527 coeff[i] = inbuf[i] * fac;
528 fac *= gamma;
529 }
530 }
531
532 static void residual_filter(float *output, const float *input,
533 const float *coef, float *memory, int length)
534 {
535 float sum;
536 int i, j;
537
538 for (i = 0; i < length; i++) {
539 sum = input[i];
540
541 for (j = FILTER_ORDER - 1; j > 0; j--) {
542 sum += coef[j] * memory[j];
543 memory[j] = memory[j - 1];
544 }
545 sum += coef[0] * memory[0];
546 memory[0] = input[i];
547 output[i] = sum;
548 }
549 }
550
551 /*
552 * TIA/IS-127 Table 5.9.1-1.
553 */
554 static const struct PfCoeff {
555 float tilt;
556 float ltgain;
557 float p1;
558 float p2;
559 } postfilter_coeffs[5] = {
560 { 0.0 , 0.0 , 0.0 , 0.0 },
561 { 0.0 , 0.0 , 0.57, 0.57 },
562 { 0.0 , 0.0 , 0.0 , 0.0 },
563 { 0.35, 0.50, 0.50, 0.75 },
564 { 0.20, 0.50, 0.57, 0.75 },
565 };
566
567 /*
568 * Adaptive postfilter.
569 *
570 * TIA/IS-127 5.9
571 */
572 static void postfilter(EVRCContext *e, float *in, const float *coeff,
573 float *out, int idx, const struct PfCoeff *pfc,
574 int length)
575 {
576 float wcoef1[FILTER_ORDER], wcoef2[FILTER_ORDER],
577 scratch[SUBFRAME_SIZE], temp[SUBFRAME_SIZE],
578 mem[SUBFRAME_SIZE];
579 float sum1 = 0.0, sum2 = 0.0, gamma, gain;
580 float tilt = pfc->tilt;
581 int i, n, best;
582
583 bandwidth_expansion(wcoef1, coeff, pfc->p1);
584 bandwidth_expansion(wcoef2, coeff, pfc->p2);
585
586 /* Tilt compensation filter, TIA/IS-127 5.9.1 */
587 for (i = 0; i < length - 1; i++)
588 sum2 += in[i] * in[i + 1];
589 if (sum2 < 0.0)
590 tilt = 0.0;
591
592 for (i = 0; i < length; i++) {
593 scratch[i] = in[i] - tilt * e->last;
594 e->last = in[i];
595 }
596
597 /* Short term residual filter, TIA/IS-127 5.9.2 */
598 residual_filter(&e->postfilter_residual[ACB_SIZE], scratch, wcoef1, e->postfilter_fir, length);
599
600 /* Long term postfilter */
601 best = idx;
602 for (i = FFMIN(MIN_DELAY, idx - 3); i <= FFMAX(MAX_DELAY, idx + 3); i++) {
603 for (n = ACB_SIZE, sum2 = 0; n < ACB_SIZE + length; n++)
604 sum2 += e->postfilter_residual[n] * e->postfilter_residual[n - i];
605 if (sum2 > sum1) {
606 sum1 = sum2;
607 best = i;
608 }
609 }
610
611 for (i = ACB_SIZE, sum1 = 0; i < ACB_SIZE + length; i++)
612 sum1 += e->postfilter_residual[i - best] * e->postfilter_residual[i - best];
613 for (i = ACB_SIZE, sum2 = 0; i < ACB_SIZE + length; i++)
614 sum2 += e->postfilter_residual[i] * e->postfilter_residual[i - best];
615
616 if (sum2 * sum1 == 0 || e->bitrate == RATE_QUANT) {
617 memcpy(temp, e->postfilter_residual + ACB_SIZE, length * sizeof(float));
618 } else {
619 gamma = sum2 / sum1;
620 if (gamma < 0.5)
621 memcpy(temp, e->postfilter_residual + ACB_SIZE, length * sizeof(float));
622 else {
623 gamma = FFMIN(gamma, 1.0);
624
625 for (i = 0; i < length; i++) {
626 temp[i] = e->postfilter_residual[ACB_SIZE + i] + gamma *
627 pfc->ltgain * e->postfilter_residual[ACB_SIZE + i - best];
628 }
629 }
630 }
631
632 memcpy(scratch, temp, length * sizeof(float));
633 memcpy(mem, e->postfilter_iir, FILTER_ORDER * sizeof(float));
634 synthesis_filter(scratch, wcoef2, mem, length, scratch);
635
636 /* Gain computation, TIA/IS-127 5.9.4-2 */
637 for (i = 0, sum1 = 0, sum2 = 0; i < length; i++) {
638 sum1 += in[i] * in[i];
639 sum2 += scratch[i] * scratch[i];
640 }
641 gain = sum2 ? sqrt(sum1 / sum2) : 1.0;
642
643 for (i = 0; i < length; i++)
644 temp[i] *= gain;
645
646 /* Short term postfilter */
647 synthesis_filter(temp, wcoef2, e->postfilter_iir, length, out);
648
649 memmove(e->postfilter_residual,
650 e->postfilter_residual + length, ACB_SIZE * sizeof(float));
651 }
652
653 static void frame_erasure(EVRCContext *e, float *samples)
654 {
655 float ilspf[FILTER_ORDER], ilpc[FILTER_ORDER], idelay[NB_SUBFRAMES],
656 tmp[SUBFRAME_SIZE + 6], f;
657 int i, j;
658
659 for (i = 0; i < FILTER_ORDER; i++) {
660 if (e->bitrate != RATE_QUANT)
661 e->lspf[i] = e->prev_lspf[i] * 0.875 + 0.125 * (i + 1) * 0.048;
662 else
663 e->lspf[i] = e->prev_lspf[i];
664 }
665
666 if (e->prev_error_flag)
667 e->avg_acb_gain *= 0.75;
668 if (e->bitrate == RATE_FULL)
669 memcpy(e->pitch_back, e->pitch, ACB_SIZE * sizeof(float));
670 if (e->last_valid_bitrate == RATE_QUANT)
671 e->bitrate = RATE_QUANT;
672 else
673 e->bitrate = RATE_FULL;
674
675 if (e->bitrate == RATE_FULL || e->bitrate == RATE_HALF) {
676 e->pitch_delay = e->prev_pitch_delay;
677 } else {
678 float sum = 0;
679
680 idelay[0] = idelay[1] = idelay[2] = MIN_DELAY;
681
682 for (i = 0; i < NB_SUBFRAMES; i++)
683 sum += evrc_energy_quant[e->prev_energy_gain][i];
684 sum /= (float) NB_SUBFRAMES;
685 sum = pow(10, sum);
686 for (i = 0; i < NB_SUBFRAMES; i++)
687 e->energy_vector[i] = sum;
688 }
689
690 if (fabs(e->pitch_delay - e->prev_pitch_delay) > 15)
691 e->prev_pitch_delay = e->pitch_delay;
692
693 for (i = 0; i < NB_SUBFRAMES; i++) {
694 int subframe_size = subframe_sizes[i];
695 int pitch_lag;
696
697 interpolate_lsp(ilspf, e->lspf, e->prev_lspf, i);
698
699 if (e->bitrate != RATE_QUANT) {
700 if (e->avg_acb_gain < 0.3) {
701 idelay[0] = estimation_delay[i];
702 idelay[1] = estimation_delay[i + 1];
703 idelay[2] = estimation_delay[i + 2];
704 } else {
705 interpolate_delay(idelay, e->pitch_delay, e->prev_pitch_delay, i);
706 }
707 }
708
709 pitch_lag = lrintf((idelay[1] + idelay[0]) / 2.0);
710 decode_predictor_coeffs(ilspf, ilpc);
711
712 if (e->bitrate != RATE_QUANT) {
713 acb_excitation(e, e->pitch + ACB_SIZE,
714 e->avg_acb_gain, idelay, subframe_size);
715 for (j = 0; j < subframe_size; j++)
716 e->pitch[ACB_SIZE + j] *= e->fade_scale;
717 e->fade_scale = FFMAX(e->fade_scale - 0.05, 0.0);
718 } else {
719 for (j = 0; j < subframe_size; j++)
720 e->pitch[ACB_SIZE + j] = e->energy_vector[i];
721 }
722
723 memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
724
725 if (e->bitrate != RATE_QUANT && e->avg_acb_gain < 0.4) {
726 f = 0.1 * e->avg_fcb_gain;
727 for (j = 0; j < subframe_size; j++)
728 e->pitch[ACB_SIZE + j] += f;
729 } else if (e->bitrate == RATE_QUANT) {
730 for (j = 0; j < subframe_size; j++)
731 e->pitch[ACB_SIZE + j] = e->energy_vector[i];
732 }
733
734 synthesis_filter(e->pitch + ACB_SIZE, ilpc,
735 e->synthesis, subframe_size, tmp);
736 postfilter(e, tmp, ilpc, samples, pitch_lag,
737 &postfilter_coeffs[e->bitrate], subframe_size);
738
739 samples += subframe_size;
740 }
741 }
742
743 static int evrc_decode_frame(AVCodecContext *avctx, void *data,
744 int *got_frame_ptr, AVPacket *avpkt)
745 {
746 const uint8_t *buf = avpkt->data;
747 AVFrame *frame = data;
748 EVRCContext *e = avctx->priv_data;
749 int buf_size = avpkt->size;
750 float ilspf[FILTER_ORDER], ilpc[FILTER_ORDER], idelay[NB_SUBFRAMES];
751 float *samples;
752 int i, j, ret, error_flag = 0;
753
754 frame->nb_samples = 160;
755 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
756 return ret;
757 samples = (float *)frame->data[0];
758
759 if ((e->bitrate = determine_bitrate(avctx, &buf_size, &buf)) == RATE_ERRS) {
760 warn_insufficient_frame_quality(avctx, "bitrate cannot be determined.");
761 goto erasure;
762 }
763 if (e->bitrate <= SILENCE || e->bitrate == RATE_QUARTER)
764 goto erasure;
765 if (e->bitrate == RATE_QUANT && e->last_valid_bitrate == RATE_FULL
766 && !e->prev_error_flag)
767 goto erasure;
768
769 if ((ret = init_get_bits8(&e->gb, buf, buf_size)) < 0)
770 return ret;
771 memset(&e->frame, 0, sizeof(EVRCAFrame));
772
773 unpack_frame(e);
774
775 if (e->bitrate != RATE_QUANT) {
776 uint8_t *p = (uint8_t *) &e->frame;
777 for (i = 0; i < sizeof(EVRCAFrame); i++) {
778 if (p[i])
779 break;
780 }
781 if (i == sizeof(EVRCAFrame))
782 goto erasure;
783 } else if (e->frame.lsp[0] == 0xf &&
784 e->frame.lsp[1] == 0xf &&
785 e->frame.energy_gain == 0xff) {
786 goto erasure;
787 }
788
789 if (decode_lspf(e) < 0)
790 goto erasure;
791
792 if (e->bitrate == RATE_FULL || e->bitrate == RATE_HALF) {
793 /* Pitch delay parameter checking as per TIA/IS-127 5.1.5.1 */
794 if (e->frame.pitch_delay > MAX_DELAY - MIN_DELAY)
795 goto erasure;
796
797 e->pitch_delay = e->frame.pitch_delay + MIN_DELAY;
798
799 /* Delay diff parameter checking as per TIA/IS-127 5.1.5.2 */
800 if (e->frame.delay_diff) {
801 int p = e->pitch_delay - e->frame.delay_diff + 16;
802 if (p < MIN_DELAY || p > MAX_DELAY)
803 goto erasure;
804 }
805
806 /* Delay contour reconstruction as per TIA/IS-127 5.2.2.2 */
807 if (e->frame.delay_diff &&
808 e->bitrate == RATE_FULL && e->prev_error_flag) {
809 float delay;
810
811 memcpy(e->pitch, e->pitch_back, ACB_SIZE * sizeof(float));
812
813 delay = e->prev_pitch_delay;
814 e->prev_pitch_delay = delay - e->frame.delay_diff + 16.0;
815
816 if (fabs(e->pitch_delay - delay) > 15)
817 delay = e->pitch_delay;
818
819 for (i = 0; i < NB_SUBFRAMES; i++) {
820 int subframe_size = subframe_sizes[i];
821
822 interpolate_delay(idelay, delay, e->prev_pitch_delay, i);
823 acb_excitation(e, e->pitch + ACB_SIZE, e->avg_acb_gain, idelay, subframe_size);
824 memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
825 }
826 }
827
828 /* Smoothing of the decoded delay as per TIA/IS-127 5.2.2.5 */
829 if (fabs(e->pitch_delay - e->prev_pitch_delay) > 15)
830 e->prev_pitch_delay = e->pitch_delay;
831
832 e->avg_acb_gain = e->avg_fcb_gain = 0.0;
833 } else {
834 idelay[0] = idelay[1] = idelay[2] = MIN_DELAY;
835
836 /* Decode frame energy vectors as per TIA/IS-127 5.7.2 */
837 for (i = 0; i < NB_SUBFRAMES; i++)
838 e->energy_vector[i] = pow(10, evrc_energy_quant[e->frame.energy_gain][i]);
839 e->prev_energy_gain = e->frame.energy_gain;
840 }
841
842 for (i = 0; i < NB_SUBFRAMES; i++) {
843 float tmp[SUBFRAME_SIZE + 6] = { 0 };
844 int subframe_size = subframe_sizes[i];
845 int pitch_lag;
846
847 interpolate_lsp(ilspf, e->lspf, e->prev_lspf, i);
848
849 if (e->bitrate != RATE_QUANT)
850 interpolate_delay(idelay, e->pitch_delay, e->prev_pitch_delay, i);
851
852 pitch_lag = lrintf((idelay[1] + idelay[0]) / 2.0);
853 decode_predictor_coeffs(ilspf, ilpc);
854
855 /* Bandwidth expansion as per TIA/IS-127 5.2.3.3 */
856 if (e->frame.lpc_flag && e->prev_error_flag)
857 bandwidth_expansion(ilpc, ilpc, 0.75);
858
859 if (e->bitrate != RATE_QUANT) {
860 float acb_sum, f;
861
862 f = exp((e->bitrate == RATE_HALF ? 0.5 : 0.25)
863 * (e->frame.fcb_gain[i] + 1));
864 acb_sum = pitch_gain_vq[e->frame.acb_gain[i]];
865 e->avg_acb_gain += acb_sum / NB_SUBFRAMES;
866 e->avg_fcb_gain += f / NB_SUBFRAMES;
867
868 acb_excitation(e, e->pitch + ACB_SIZE,
869 acb_sum, idelay, subframe_size);
870 fcb_excitation(e, e->frame.fcb_shape[i], tmp,
871 acb_sum, pitch_lag, subframe_size);
872
873 /* Total excitation generation as per TIA/IS-127 5.2.3.9 */
874 for (j = 0; j < subframe_size; j++)
875 e->pitch[ACB_SIZE + j] += f * tmp[j];
876 e->fade_scale = FFMIN(e->fade_scale + 0.2, 1.0);
877 } else {
878 for (j = 0; j < subframe_size; j++)
879 e->pitch[ACB_SIZE + j] = e->energy_vector[i];
880 }
881
882 memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
883
884 synthesis_filter(e->pitch + ACB_SIZE, ilpc,
885 e->synthesis, subframe_size,
886 e->postfilter ? tmp : samples);
887 if (e->postfilter)
888 postfilter(e, tmp, ilpc, samples, pitch_lag,
889 &postfilter_coeffs[e->bitrate], subframe_size);
890
891 samples += subframe_size;
892 }
893
894 if (error_flag) {
895 erasure:
896 error_flag = 1;
897 av_log(avctx, AV_LOG_WARNING, "frame erasure\n");
898 frame_erasure(e, samples);
899 }
900
901 memcpy(e->prev_lspf, e->lspf, sizeof(e->prev_lspf));
902 e->prev_error_flag = error_flag;
903 e->last_valid_bitrate = e->bitrate;
904
905 if (e->bitrate != RATE_QUANT)
906 e->prev_pitch_delay = e->pitch_delay;
907
908 samples = (float *)frame->data[0];
909 for (i = 0; i < 160; i++)
910 samples[i] /= 32768;
911
912 *got_frame_ptr = 1;
913
914 return avpkt->size;
915 }
916
917 #define OFFSET(x) offsetof(EVRCContext, x)
918 #define AD AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM
919
920 static const AVOption options[] = {
921 { "postfilter", "enable postfilter", OFFSET(postfilter), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0, 1, AD },
922 { NULL }
923 };
924
925 static const AVClass evrcdec_class = {
926 .class_name = "evrc",
927 .item_name = av_default_item_name,
928 .option = options,
929 .version = LIBAVUTIL_VERSION_INT,
930 };
931
932 const AVCodec ff_evrc_decoder = {
933 .name = "evrc",
934 .long_name = NULL_IF_CONFIG_SMALL("EVRC (Enhanced Variable Rate Codec)"),
935 .type = AVMEDIA_TYPE_AUDIO,
936 .id = AV_CODEC_ID_EVRC,
937 .init = evrc_decode_init,
938 .decode = evrc_decode_frame,
939 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
940 .priv_data_size = sizeof(EVRCContext),
941 .priv_class = &evrcdec_class,
942 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
943 };
944