3 * Copyright (c) 2010 Marcelo Galvao Povoa
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * AMR wideband decoder
27 #include "libavutil/channel_layout.h"
28 #include "libavutil/common.h"
29 #include "libavutil/lfg.h"
34 #include "celp_filters.h"
35 #include "acelp_filters.h"
36 #include "acelp_vectors.h"
37 #include "acelp_pitch_delay.h"
40 #define AMR_USE_16BIT_TABLES
43 #include "amrwbdata.h"
46 AVFrame avframe; ///< AVFrame for decoded samples
47 AMRWBFrame frame; ///< AMRWB parameters decoded from bitstream
48 enum Mode fr_cur_mode; ///< mode index of current frame
49 uint8_t fr_quality; ///< frame quality index (FQI)
50 float isf_cur[LP_ORDER]; ///< working ISF vector from current frame
51 float isf_q_past[LP_ORDER]; ///< quantized ISF vector of the previous frame
52 float isf_past_final[LP_ORDER]; ///< final processed ISF vector of the previous frame
53 double isp[4][LP_ORDER]; ///< ISP vectors from current frame
54 double isp_sub4_past[LP_ORDER]; ///< ISP vector for the 4th subframe of the previous frame
56 float lp_coef[4][LP_ORDER]; ///< Linear Prediction Coefficients from ISP vector
58 uint8_t base_pitch_lag; ///< integer part of pitch lag for the next relative subframe
59 uint8_t pitch_lag_int; ///< integer part of pitch lag of the previous subframe
61 float excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 2 + AMRWB_SFR_SIZE]; ///< current excitation and all necessary excitation history
62 float *excitation; ///< points to current excitation in excitation_buf[]
64 float pitch_vector[AMRWB_SFR_SIZE]; ///< adaptive codebook (pitch) vector for current subframe
65 float fixed_vector[AMRWB_SFR_SIZE]; ///< algebraic codebook (fixed) vector for current subframe
67 float prediction_error[4]; ///< quantified prediction errors {20log10(^gamma_gc)} for previous four subframes
68 float pitch_gain[6]; ///< quantified pitch gains for the current and previous five subframes
69 float fixed_gain[2]; ///< quantified fixed gains for the current and previous subframes
71 float tilt_coef; ///< {beta_1} related to the voicing of the previous subframe
73 float prev_sparse_fixed_gain; ///< previous fixed gain; used by anti-sparseness to determine "onset"
74 uint8_t prev_ir_filter_nr; ///< previous impulse response filter "impNr": 0 - strong, 1 - medium, 2 - none
75 float prev_tr_gain; ///< previous initial gain used by noise enhancer for threshold
77 float samples_az[LP_ORDER + AMRWB_SFR_SIZE]; ///< low-band samples and memory from synthesis at 12.8kHz
78 float samples_up[UPS_MEM_SIZE + AMRWB_SFR_SIZE]; ///< low-band samples and memory processed for upsampling
79 float samples_hb[LP_ORDER_16k + AMRWB_SFR_SIZE_16k]; ///< high-band samples and memory from synthesis at 16kHz
81 float hpf_31_mem[2], hpf_400_mem[2]; ///< previous values in the high pass filters
82 float demph_mem[1]; ///< previous value in the de-emphasis filter
83 float bpf_6_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band band pass filter
84 float lpf_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band low pass filter
86 AVLFG prng; ///< random number generator for white noise excitation
87 uint8_t first_frame; ///< flag active during decoding of the first frame
90 static av_cold int amrwb_decode_init(AVCodecContext *avctx)
92 AMRWBContext *ctx = avctx->priv_data;
95 if (avctx->channels > 1) {
96 av_log_missing_feature(avctx, "multi-channel AMR", 0);
97 return AVERROR_PATCHWELCOME;
101 avctx->channel_layout = AV_CH_LAYOUT_MONO;
102 avctx->sample_rate = 16000;
103 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
105 av_lfg_init(&ctx->prng, 1);
107 ctx->excitation = &ctx->excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 1];
108 ctx->first_frame = 1;
110 for (i = 0; i < LP_ORDER; i++)
111 ctx->isf_past_final[i] = isf_init[i] * (1.0f / (1 << 15));
113 for (i = 0; i < 4; i++)
114 ctx->prediction_error[i] = MIN_ENERGY;
116 avcodec_get_frame_defaults(&ctx->avframe);
117 avctx->coded_frame = &ctx->avframe;
123 * Decode the frame header in the "MIME/storage" format. This format
124 * is simpler and does not carry the auxiliary frame information.
126 * @param[in] ctx The Context
127 * @param[in] buf Pointer to the input buffer
129 * @return The decoded header length in bytes
131 static int decode_mime_header(AMRWBContext *ctx, const uint8_t *buf)
133 /* Decode frame header (1st octet) */
134 ctx->fr_cur_mode = buf[0] >> 3 & 0x0F;
135 ctx->fr_quality = (buf[0] & 0x4) == 0x4;
141 * Decode quantized ISF vectors using 36-bit indexes (6K60 mode only).
143 * @param[in] ind Array of 5 indexes
144 * @param[out] isf_q Buffer for isf_q[LP_ORDER]
147 static void decode_isf_indices_36b(uint16_t *ind, float *isf_q)
151 for (i = 0; i < 9; i++)
152 isf_q[i] = dico1_isf[ind[0]][i] * (1.0f / (1 << 15));
154 for (i = 0; i < 7; i++)
155 isf_q[i + 9] = dico2_isf[ind[1]][i] * (1.0f / (1 << 15));
157 for (i = 0; i < 5; i++)
158 isf_q[i] += dico21_isf_36b[ind[2]][i] * (1.0f / (1 << 15));
160 for (i = 0; i < 4; i++)
161 isf_q[i + 5] += dico22_isf_36b[ind[3]][i] * (1.0f / (1 << 15));
163 for (i = 0; i < 7; i++)
164 isf_q[i + 9] += dico23_isf_36b[ind[4]][i] * (1.0f / (1 << 15));
168 * Decode quantized ISF vectors using 46-bit indexes (except 6K60 mode).
170 * @param[in] ind Array of 7 indexes
171 * @param[out] isf_q Buffer for isf_q[LP_ORDER]
174 static void decode_isf_indices_46b(uint16_t *ind, float *isf_q)
178 for (i = 0; i < 9; i++)
179 isf_q[i] = dico1_isf[ind[0]][i] * (1.0f / (1 << 15));
181 for (i = 0; i < 7; i++)
182 isf_q[i + 9] = dico2_isf[ind[1]][i] * (1.0f / (1 << 15));
184 for (i = 0; i < 3; i++)
185 isf_q[i] += dico21_isf[ind[2]][i] * (1.0f / (1 << 15));
187 for (i = 0; i < 3; i++)
188 isf_q[i + 3] += dico22_isf[ind[3]][i] * (1.0f / (1 << 15));
190 for (i = 0; i < 3; i++)
191 isf_q[i + 6] += dico23_isf[ind[4]][i] * (1.0f / (1 << 15));
193 for (i = 0; i < 3; i++)
194 isf_q[i + 9] += dico24_isf[ind[5]][i] * (1.0f / (1 << 15));
196 for (i = 0; i < 4; i++)
197 isf_q[i + 12] += dico25_isf[ind[6]][i] * (1.0f / (1 << 15));
201 * Apply mean and past ISF values using the prediction factor.
202 * Updates past ISF vector.
204 * @param[in,out] isf_q Current quantized ISF
205 * @param[in,out] isf_past Past quantized ISF
208 static void isf_add_mean_and_past(float *isf_q, float *isf_past)
213 for (i = 0; i < LP_ORDER; i++) {
215 isf_q[i] += isf_mean[i] * (1.0f / (1 << 15));
216 isf_q[i] += PRED_FACTOR * isf_past[i];
222 * Interpolate the fourth ISP vector from current and past frames
223 * to obtain an ISP vector for each subframe.
225 * @param[in,out] isp_q ISPs for each subframe
226 * @param[in] isp4_past Past ISP for subframe 4
228 static void interpolate_isp(double isp_q[4][LP_ORDER], const double *isp4_past)
232 for (k = 0; k < 3; k++) {
233 float c = isfp_inter[k];
234 for (i = 0; i < LP_ORDER; i++)
235 isp_q[k][i] = (1.0 - c) * isp4_past[i] + c * isp_q[3][i];
240 * Decode an adaptive codebook index into pitch lag (except 6k60, 8k85 modes).
241 * Calculate integer lag and fractional lag always using 1/4 resolution.
242 * In 1st and 3rd subframes the index is relative to last subframe integer lag.
244 * @param[out] lag_int Decoded integer pitch lag
245 * @param[out] lag_frac Decoded fractional pitch lag
246 * @param[in] pitch_index Adaptive codebook pitch index
247 * @param[in,out] base_lag_int Base integer lag used in relative subframes
248 * @param[in] subframe Current subframe index (0 to 3)
250 static void decode_pitch_lag_high(int *lag_int, int *lag_frac, int pitch_index,
251 uint8_t *base_lag_int, int subframe)
253 if (subframe == 0 || subframe == 2) {
254 if (pitch_index < 376) {
255 *lag_int = (pitch_index + 137) >> 2;
256 *lag_frac = pitch_index - (*lag_int << 2) + 136;
257 } else if (pitch_index < 440) {
258 *lag_int = (pitch_index + 257 - 376) >> 1;
259 *lag_frac = (pitch_index - (*lag_int << 1) + 256 - 376) << 1;
260 /* the actual resolution is 1/2 but expressed as 1/4 */
262 *lag_int = pitch_index - 280;
265 /* minimum lag for next subframe */
266 *base_lag_int = av_clip(*lag_int - 8 - (*lag_frac < 0),
267 AMRWB_P_DELAY_MIN, AMRWB_P_DELAY_MAX - 15);
268 // XXX: the spec states clearly that *base_lag_int should be
269 // the nearest integer to *lag_int (minus 8), but the ref code
270 // actually always uses its floor, I'm following the latter
272 *lag_int = (pitch_index + 1) >> 2;
273 *lag_frac = pitch_index - (*lag_int << 2);
274 *lag_int += *base_lag_int;
279 * Decode an adaptive codebook index into pitch lag for 8k85 and 6k60 modes.
280 * The description is analogous to decode_pitch_lag_high, but in 6k60 the
281 * relative index is used for all subframes except the first.
283 static void decode_pitch_lag_low(int *lag_int, int *lag_frac, int pitch_index,
284 uint8_t *base_lag_int, int subframe, enum Mode mode)
286 if (subframe == 0 || (subframe == 2 && mode != MODE_6k60)) {
287 if (pitch_index < 116) {
288 *lag_int = (pitch_index + 69) >> 1;
289 *lag_frac = (pitch_index - (*lag_int << 1) + 68) << 1;
291 *lag_int = pitch_index - 24;
294 // XXX: same problem as before
295 *base_lag_int = av_clip(*lag_int - 8 - (*lag_frac < 0),
296 AMRWB_P_DELAY_MIN, AMRWB_P_DELAY_MAX - 15);
298 *lag_int = (pitch_index + 1) >> 1;
299 *lag_frac = (pitch_index - (*lag_int << 1)) << 1;
300 *lag_int += *base_lag_int;
305 * Find the pitch vector by interpolating the past excitation at the
306 * pitch delay, which is obtained in this function.
308 * @param[in,out] ctx The context
309 * @param[in] amr_subframe Current subframe data
310 * @param[in] subframe Current subframe index (0 to 3)
312 static void decode_pitch_vector(AMRWBContext *ctx,
313 const AMRWBSubFrame *amr_subframe,
316 int pitch_lag_int, pitch_lag_frac;
318 float *exc = ctx->excitation;
319 enum Mode mode = ctx->fr_cur_mode;
321 if (mode <= MODE_8k85) {
322 decode_pitch_lag_low(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap,
323 &ctx->base_pitch_lag, subframe, mode);
325 decode_pitch_lag_high(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap,
326 &ctx->base_pitch_lag, subframe);
328 ctx->pitch_lag_int = pitch_lag_int;
329 pitch_lag_int += pitch_lag_frac > 0;
331 /* Calculate the pitch vector by interpolating the past excitation at the
332 pitch lag using a hamming windowed sinc function */
333 ff_acelp_interpolatef(exc, exc + 1 - pitch_lag_int,
335 pitch_lag_frac + (pitch_lag_frac > 0 ? 0 : 4),
336 LP_ORDER, AMRWB_SFR_SIZE + 1);
338 /* Check which pitch signal path should be used
339 * 6k60 and 8k85 modes have the ltp flag set to 0 */
340 if (amr_subframe->ltp) {
341 memcpy(ctx->pitch_vector, exc, AMRWB_SFR_SIZE * sizeof(float));
343 for (i = 0; i < AMRWB_SFR_SIZE; i++)
344 ctx->pitch_vector[i] = 0.18 * exc[i - 1] + 0.64 * exc[i] +
346 memcpy(exc, ctx->pitch_vector, AMRWB_SFR_SIZE * sizeof(float));
350 /** Get x bits in the index interval [lsb,lsb+len-1] inclusive */
351 #define BIT_STR(x,lsb,len) (((x) >> (lsb)) & ((1 << (len)) - 1))
353 /** Get the bit at specified position */
354 #define BIT_POS(x, p) (((x) >> (p)) & 1)
357 * The next six functions decode_[i]p_track decode exactly i pulses
358 * positions and amplitudes (-1 or 1) in a subframe track using
359 * an encoded pulse indexing (TS 26.190 section 5.8.2).
361 * The results are given in out[], in which a negative number means
362 * amplitude -1 and vice versa (i.e., ampl(x) = x / abs(x) ).
364 * @param[out] out Output buffer (writes i elements)
365 * @param[in] code Pulse index (no. of bits varies, see below)
366 * @param[in] m (log2) Number of potential positions
367 * @param[in] off Offset for decoded positions
369 static inline void decode_1p_track(int *out, int code, int m, int off)
371 int pos = BIT_STR(code, 0, m) + off; ///code: m+1 bits
373 out[0] = BIT_POS(code, m) ? -pos : pos;
376 static inline void decode_2p_track(int *out, int code, int m, int off) ///code: 2m+1 bits
378 int pos0 = BIT_STR(code, m, m) + off;
379 int pos1 = BIT_STR(code, 0, m) + off;
381 out[0] = BIT_POS(code, 2*m) ? -pos0 : pos0;
382 out[1] = BIT_POS(code, 2*m) ? -pos1 : pos1;
383 out[1] = pos0 > pos1 ? -out[1] : out[1];
386 static void decode_3p_track(int *out, int code, int m, int off) ///code: 3m+1 bits
388 int half_2p = BIT_POS(code, 2*m - 1) << (m - 1);
390 decode_2p_track(out, BIT_STR(code, 0, 2*m - 1),
391 m - 1, off + half_2p);
392 decode_1p_track(out + 2, BIT_STR(code, 2*m, m + 1), m, off);
395 static void decode_4p_track(int *out, int code, int m, int off) ///code: 4m bits
397 int half_4p, subhalf_2p;
398 int b_offset = 1 << (m - 1);
400 switch (BIT_STR(code, 4*m - 2, 2)) { /* case ID (2 bits) */
401 case 0: /* 0 pulses in A, 4 pulses in B or vice versa */
402 half_4p = BIT_POS(code, 4*m - 3) << (m - 1); // which has 4 pulses
403 subhalf_2p = BIT_POS(code, 2*m - 3) << (m - 2);
405 decode_2p_track(out, BIT_STR(code, 0, 2*m - 3),
406 m - 2, off + half_4p + subhalf_2p);
407 decode_2p_track(out + 2, BIT_STR(code, 2*m - 2, 2*m - 1),
408 m - 1, off + half_4p);
410 case 1: /* 1 pulse in A, 3 pulses in B */
411 decode_1p_track(out, BIT_STR(code, 3*m - 2, m),
413 decode_3p_track(out + 1, BIT_STR(code, 0, 3*m - 2),
414 m - 1, off + b_offset);
416 case 2: /* 2 pulses in each half */
417 decode_2p_track(out, BIT_STR(code, 2*m - 1, 2*m - 1),
419 decode_2p_track(out + 2, BIT_STR(code, 0, 2*m - 1),
420 m - 1, off + b_offset);
422 case 3: /* 3 pulses in A, 1 pulse in B */
423 decode_3p_track(out, BIT_STR(code, m, 3*m - 2),
425 decode_1p_track(out + 3, BIT_STR(code, 0, m),
426 m - 1, off + b_offset);
431 static void decode_5p_track(int *out, int code, int m, int off) ///code: 5m bits
433 int half_3p = BIT_POS(code, 5*m - 1) << (m - 1);
435 decode_3p_track(out, BIT_STR(code, 2*m + 1, 3*m - 2),
436 m - 1, off + half_3p);
438 decode_2p_track(out + 3, BIT_STR(code, 0, 2*m + 1), m, off);
441 static void decode_6p_track(int *out, int code, int m, int off) ///code: 6m-2 bits
443 int b_offset = 1 << (m - 1);
444 /* which half has more pulses in cases 0 to 2 */
445 int half_more = BIT_POS(code, 6*m - 5) << (m - 1);
446 int half_other = b_offset - half_more;
448 switch (BIT_STR(code, 6*m - 4, 2)) { /* case ID (2 bits) */
449 case 0: /* 0 pulses in A, 6 pulses in B or vice versa */
450 decode_1p_track(out, BIT_STR(code, 0, m),
451 m - 1, off + half_more);
452 decode_5p_track(out + 1, BIT_STR(code, m, 5*m - 5),
453 m - 1, off + half_more);
455 case 1: /* 1 pulse in A, 5 pulses in B or vice versa */
456 decode_1p_track(out, BIT_STR(code, 0, m),
457 m - 1, off + half_other);
458 decode_5p_track(out + 1, BIT_STR(code, m, 5*m - 5),
459 m - 1, off + half_more);
461 case 2: /* 2 pulses in A, 4 pulses in B or vice versa */
462 decode_2p_track(out, BIT_STR(code, 0, 2*m - 1),
463 m - 1, off + half_other);
464 decode_4p_track(out + 2, BIT_STR(code, 2*m - 1, 4*m - 4),
465 m - 1, off + half_more);
467 case 3: /* 3 pulses in A, 3 pulses in B */
468 decode_3p_track(out, BIT_STR(code, 3*m - 2, 3*m - 2),
470 decode_3p_track(out + 3, BIT_STR(code, 0, 3*m - 2),
471 m - 1, off + b_offset);
477 * Decode the algebraic codebook index to pulse positions and signs,
478 * then construct the algebraic codebook vector.
480 * @param[out] fixed_vector Buffer for the fixed codebook excitation
481 * @param[in] pulse_hi MSBs part of the pulse index array (higher modes only)
482 * @param[in] pulse_lo LSBs part of the pulse index array
483 * @param[in] mode Mode of the current frame
485 static void decode_fixed_vector(float *fixed_vector, const uint16_t *pulse_hi,
486 const uint16_t *pulse_lo, const enum Mode mode)
488 /* sig_pos stores for each track the decoded pulse position indexes
489 * (1-based) multiplied by its corresponding amplitude (+1 or -1) */
491 int spacing = (mode == MODE_6k60) ? 2 : 4;
496 for (i = 0; i < 2; i++)
497 decode_1p_track(sig_pos[i], pulse_lo[i], 5, 1);
500 for (i = 0; i < 4; i++)
501 decode_1p_track(sig_pos[i], pulse_lo[i], 4, 1);
504 for (i = 0; i < 4; i++)
505 decode_2p_track(sig_pos[i], pulse_lo[i], 4, 1);
508 for (i = 0; i < 2; i++)
509 decode_3p_track(sig_pos[i], pulse_lo[i], 4, 1);
510 for (i = 2; i < 4; i++)
511 decode_2p_track(sig_pos[i], pulse_lo[i], 4, 1);
514 for (i = 0; i < 4; i++)
515 decode_3p_track(sig_pos[i], pulse_lo[i], 4, 1);
518 for (i = 0; i < 4; i++)
519 decode_4p_track(sig_pos[i], (int) pulse_lo[i] +
520 ((int) pulse_hi[i] << 14), 4, 1);
523 for (i = 0; i < 2; i++)
524 decode_5p_track(sig_pos[i], (int) pulse_lo[i] +
525 ((int) pulse_hi[i] << 10), 4, 1);
526 for (i = 2; i < 4; i++)
527 decode_4p_track(sig_pos[i], (int) pulse_lo[i] +
528 ((int) pulse_hi[i] << 14), 4, 1);
532 for (i = 0; i < 4; i++)
533 decode_6p_track(sig_pos[i], (int) pulse_lo[i] +
534 ((int) pulse_hi[i] << 11), 4, 1);
538 memset(fixed_vector, 0, sizeof(float) * AMRWB_SFR_SIZE);
540 for (i = 0; i < 4; i++)
541 for (j = 0; j < pulses_nb_per_mode_tr[mode][i]; j++) {
542 int pos = (FFABS(sig_pos[i][j]) - 1) * spacing + i;
544 fixed_vector[pos] += sig_pos[i][j] < 0 ? -1.0 : 1.0;
549 * Decode pitch gain and fixed gain correction factor.
551 * @param[in] vq_gain Vector-quantized index for gains
552 * @param[in] mode Mode of the current frame
553 * @param[out] fixed_gain_factor Decoded fixed gain correction factor
554 * @param[out] pitch_gain Decoded pitch gain
556 static void decode_gains(const uint8_t vq_gain, const enum Mode mode,
557 float *fixed_gain_factor, float *pitch_gain)
559 const int16_t *gains = (mode <= MODE_8k85 ? qua_gain_6b[vq_gain] :
560 qua_gain_7b[vq_gain]);
562 *pitch_gain = gains[0] * (1.0f / (1 << 14));
563 *fixed_gain_factor = gains[1] * (1.0f / (1 << 11));
567 * Apply pitch sharpening filters to the fixed codebook vector.
569 * @param[in] ctx The context
570 * @param[in,out] fixed_vector Fixed codebook excitation
572 // XXX: Spec states this procedure should be applied when the pitch
573 // lag is less than 64, but this checking seems absent in reference and AMR-NB
574 static void pitch_sharpening(AMRWBContext *ctx, float *fixed_vector)
579 for (i = AMRWB_SFR_SIZE - 1; i != 0; i--)
580 fixed_vector[i] -= fixed_vector[i - 1] * ctx->tilt_coef;
582 /* Periodicity enhancement part */
583 for (i = ctx->pitch_lag_int; i < AMRWB_SFR_SIZE; i++)
584 fixed_vector[i] += fixed_vector[i - ctx->pitch_lag_int] * 0.85;
588 * Calculate the voicing factor (-1.0 = unvoiced to 1.0 = voiced).
590 * @param[in] p_vector, f_vector Pitch and fixed excitation vectors
591 * @param[in] p_gain, f_gain Pitch and fixed gains
593 // XXX: There is something wrong with the precision here! The magnitudes
594 // of the energies are not correct. Please check the reference code carefully
595 static float voice_factor(float *p_vector, float p_gain,
596 float *f_vector, float f_gain)
598 double p_ener = (double) ff_scalarproduct_float_c(p_vector, p_vector,
601 double f_ener = (double) ff_scalarproduct_float_c(f_vector, f_vector,
605 return (p_ener - f_ener) / (p_ener + f_ener);
609 * Reduce fixed vector sparseness by smoothing with one of three IR filters,
610 * also known as "adaptive phase dispersion".
612 * @param[in] ctx The context
613 * @param[in,out] fixed_vector Unfiltered fixed vector
614 * @param[out] buf Space for modified vector if necessary
616 * @return The potentially overwritten filtered fixed vector address
618 static float *anti_sparseness(AMRWBContext *ctx,
619 float *fixed_vector, float *buf)
623 if (ctx->fr_cur_mode > MODE_8k85) // no filtering in higher modes
626 if (ctx->pitch_gain[0] < 0.6) {
627 ir_filter_nr = 0; // strong filtering
628 } else if (ctx->pitch_gain[0] < 0.9) {
629 ir_filter_nr = 1; // medium filtering
631 ir_filter_nr = 2; // no filtering
634 if (ctx->fixed_gain[0] > 3.0 * ctx->fixed_gain[1]) {
635 if (ir_filter_nr < 2)
640 for (i = 0; i < 6; i++)
641 if (ctx->pitch_gain[i] < 0.6)
647 if (ir_filter_nr > ctx->prev_ir_filter_nr + 1)
651 /* update ir filter strength history */
652 ctx->prev_ir_filter_nr = ir_filter_nr;
654 ir_filter_nr += (ctx->fr_cur_mode == MODE_8k85);
656 if (ir_filter_nr < 2) {
658 const float *coef = ir_filters_lookup[ir_filter_nr];
660 /* Circular convolution code in the reference
661 * decoder was modified to avoid using one
662 * extra array. The filtered vector is given by:
664 * c2(n) = sum(i,0,len-1){ c(i) * coef( (n - i + len) % len ) }
667 memset(buf, 0, sizeof(float) * AMRWB_SFR_SIZE);
668 for (i = 0; i < AMRWB_SFR_SIZE; i++)
670 ff_celp_circ_addf(buf, buf, coef, i, fixed_vector[i],
679 * Calculate a stability factor {teta} based on distance between
680 * current and past isf. A value of 1 shows maximum signal stability.
682 static float stability_factor(const float *isf, const float *isf_past)
687 for (i = 0; i < LP_ORDER - 1; i++)
688 acc += (isf[i] - isf_past[i]) * (isf[i] - isf_past[i]);
690 // XXX: This part is not so clear from the reference code
691 // the result is more accurate changing the "/ 256" to "* 512"
692 return FFMAX(0.0, 1.25 - acc * 0.8 * 512);
696 * Apply a non-linear fixed gain smoothing in order to reduce
697 * fluctuation in the energy of excitation.
699 * @param[in] fixed_gain Unsmoothed fixed gain
700 * @param[in,out] prev_tr_gain Previous threshold gain (updated)
701 * @param[in] voice_fac Frame voicing factor
702 * @param[in] stab_fac Frame stability factor
704 * @return The smoothed gain
706 static float noise_enhancer(float fixed_gain, float *prev_tr_gain,
707 float voice_fac, float stab_fac)
709 float sm_fac = 0.5 * (1 - voice_fac) * stab_fac;
712 // XXX: the following fixed-point constants used to in(de)crement
713 // gain by 1.5dB were taken from the reference code, maybe it could
715 if (fixed_gain < *prev_tr_gain) {
716 g0 = FFMIN(*prev_tr_gain, fixed_gain + fixed_gain *
717 (6226 * (1.0f / (1 << 15)))); // +1.5 dB
719 g0 = FFMAX(*prev_tr_gain, fixed_gain *
720 (27536 * (1.0f / (1 << 15)))); // -1.5 dB
722 *prev_tr_gain = g0; // update next frame threshold
724 return sm_fac * g0 + (1 - sm_fac) * fixed_gain;
728 * Filter the fixed_vector to emphasize the higher frequencies.
730 * @param[in,out] fixed_vector Fixed codebook vector
731 * @param[in] voice_fac Frame voicing factor
733 static void pitch_enhancer(float *fixed_vector, float voice_fac)
736 float cpe = 0.125 * (1 + voice_fac);
737 float last = fixed_vector[0]; // holds c(i - 1)
739 fixed_vector[0] -= cpe * fixed_vector[1];
741 for (i = 1; i < AMRWB_SFR_SIZE - 1; i++) {
742 float cur = fixed_vector[i];
744 fixed_vector[i] -= cpe * (last + fixed_vector[i + 1]);
748 fixed_vector[AMRWB_SFR_SIZE - 1] -= cpe * last;
752 * Conduct 16th order linear predictive coding synthesis from excitation.
754 * @param[in] ctx Pointer to the AMRWBContext
755 * @param[in] lpc Pointer to the LPC coefficients
756 * @param[out] excitation Buffer for synthesis final excitation
757 * @param[in] fixed_gain Fixed codebook gain for synthesis
758 * @param[in] fixed_vector Algebraic codebook vector
759 * @param[in,out] samples Pointer to the output samples and memory
761 static void synthesis(AMRWBContext *ctx, float *lpc, float *excitation,
762 float fixed_gain, const float *fixed_vector,
765 ff_weighted_vector_sumf(excitation, ctx->pitch_vector, fixed_vector,
766 ctx->pitch_gain[0], fixed_gain, AMRWB_SFR_SIZE);
768 /* emphasize pitch vector contribution in low bitrate modes */
769 if (ctx->pitch_gain[0] > 0.5 && ctx->fr_cur_mode <= MODE_8k85) {
771 float energy = ff_scalarproduct_float_c(excitation, excitation,
774 // XXX: Weird part in both ref code and spec. A unknown parameter
775 // {beta} seems to be identical to the current pitch gain
776 float pitch_factor = 0.25 * ctx->pitch_gain[0] * ctx->pitch_gain[0];
778 for (i = 0; i < AMRWB_SFR_SIZE; i++)
779 excitation[i] += pitch_factor * ctx->pitch_vector[i];
781 ff_scale_vector_to_given_sum_of_squares(excitation, excitation,
782 energy, AMRWB_SFR_SIZE);
785 ff_celp_lp_synthesis_filterf(samples, lpc, excitation,
786 AMRWB_SFR_SIZE, LP_ORDER);
790 * Apply to synthesis a de-emphasis filter of the form:
791 * H(z) = 1 / (1 - m * z^-1)
793 * @param[out] out Output buffer
794 * @param[in] in Input samples array with in[-1]
795 * @param[in] m Filter coefficient
796 * @param[in,out] mem State from last filtering
798 static void de_emphasis(float *out, float *in, float m, float mem[1])
802 out[0] = in[0] + m * mem[0];
804 for (i = 1; i < AMRWB_SFR_SIZE; i++)
805 out[i] = in[i] + out[i - 1] * m;
807 mem[0] = out[AMRWB_SFR_SIZE - 1];
811 * Upsample a signal by 5/4 ratio (from 12.8kHz to 16kHz) using
812 * a FIR interpolation filter. Uses past data from before *in address.
814 * @param[out] out Buffer for interpolated signal
815 * @param[in] in Current signal data (length 0.8*o_size)
816 * @param[in] o_size Output signal length
818 static void upsample_5_4(float *out, const float *in, int o_size)
820 const float *in0 = in - UPS_FIR_SIZE + 1;
822 int int_part = 0, frac_part;
825 for (j = 0; j < o_size / 5; j++) {
826 out[i] = in[int_part];
830 for (k = 1; k < 5; k++) {
831 out[i] = ff_scalarproduct_float_c(in0 + int_part,
832 upsample_fir[4 - frac_part],
842 * Calculate the high-band gain based on encoded index (23k85 mode) or
843 * on the low-band speech signal and the Voice Activity Detection flag.
845 * @param[in] ctx The context
846 * @param[in] synth LB speech synthesis at 12.8k
847 * @param[in] hb_idx Gain index for mode 23k85 only
848 * @param[in] vad VAD flag for the frame
850 static float find_hb_gain(AMRWBContext *ctx, const float *synth,
851 uint16_t hb_idx, uint8_t vad)
856 if (ctx->fr_cur_mode == MODE_23k85)
857 return qua_hb_gain[hb_idx] * (1.0f / (1 << 14));
859 tilt = ff_scalarproduct_float_c(synth, synth + 1, AMRWB_SFR_SIZE - 1) /
860 ff_scalarproduct_float_c(synth, synth, AMRWB_SFR_SIZE);
862 /* return gain bounded by [0.1, 1.0] */
863 return av_clipf((1.0 - FFMAX(0.0, tilt)) * (1.25 - 0.25 * wsp), 0.1, 1.0);
867 * Generate the high-band excitation with the same energy from the lower
868 * one and scaled by the given gain.
870 * @param[in] ctx The context
871 * @param[out] hb_exc Buffer for the excitation
872 * @param[in] synth_exc Low-band excitation used for synthesis
873 * @param[in] hb_gain Wanted excitation gain
875 static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc,
876 const float *synth_exc, float hb_gain)
879 float energy = ff_scalarproduct_float_c(synth_exc, synth_exc, AMRWB_SFR_SIZE);
881 /* Generate a white-noise excitation */
882 for (i = 0; i < AMRWB_SFR_SIZE_16k; i++)
883 hb_exc[i] = 32768.0 - (uint16_t) av_lfg_get(&ctx->prng);
885 ff_scale_vector_to_given_sum_of_squares(hb_exc, hb_exc,
886 energy * hb_gain * hb_gain,
891 * Calculate the auto-correlation for the ISF difference vector.
893 static float auto_correlation(float *diff_isf, float mean, int lag)
898 for (i = 7; i < LP_ORDER - 2; i++) {
899 float prod = (diff_isf[i] - mean) * (diff_isf[i - lag] - mean);
906 * Extrapolate a ISF vector to the 16kHz range (20th order LP)
907 * used at mode 6k60 LP filter for the high frequency band.
909 * @param[out] isf Buffer for extrapolated isf; contains LP_ORDER
912 static void extrapolate_isf(float isf[LP_ORDER_16k])
914 float diff_isf[LP_ORDER - 2], diff_mean;
917 int i, j, i_max_corr;
919 isf[LP_ORDER_16k - 1] = isf[LP_ORDER - 1];
921 /* Calculate the difference vector */
922 for (i = 0; i < LP_ORDER - 2; i++)
923 diff_isf[i] = isf[i + 1] - isf[i];
926 for (i = 2; i < LP_ORDER - 2; i++)
927 diff_mean += diff_isf[i] * (1.0f / (LP_ORDER - 4));
929 /* Find which is the maximum autocorrelation */
931 for (i = 0; i < 3; i++) {
932 corr_lag[i] = auto_correlation(diff_isf, diff_mean, i + 2);
934 if (corr_lag[i] > corr_lag[i_max_corr])
939 for (i = LP_ORDER - 1; i < LP_ORDER_16k - 1; i++)
940 isf[i] = isf[i - 1] + isf[i - 1 - i_max_corr]
941 - isf[i - 2 - i_max_corr];
943 /* Calculate an estimate for ISF(18) and scale ISF based on the error */
944 est = 7965 + (isf[2] - isf[3] - isf[4]) / 6.0;
945 scale = 0.5 * (FFMIN(est, 7600) - isf[LP_ORDER - 2]) /
946 (isf[LP_ORDER_16k - 2] - isf[LP_ORDER - 2]);
948 for (i = LP_ORDER - 1, j = 0; i < LP_ORDER_16k - 1; i++, j++)
949 diff_isf[j] = scale * (isf[i] - isf[i - 1]);
951 /* Stability insurance */
952 for (i = 1; i < LP_ORDER_16k - LP_ORDER; i++)
953 if (diff_isf[i] + diff_isf[i - 1] < 5.0) {
954 if (diff_isf[i] > diff_isf[i - 1]) {
955 diff_isf[i - 1] = 5.0 - diff_isf[i];
957 diff_isf[i] = 5.0 - diff_isf[i - 1];
960 for (i = LP_ORDER - 1, j = 0; i < LP_ORDER_16k - 1; i++, j++)
961 isf[i] = isf[i - 1] + diff_isf[j] * (1.0f / (1 << 15));
963 /* Scale the ISF vector for 16000 Hz */
964 for (i = 0; i < LP_ORDER_16k - 1; i++)
969 * Spectral expand the LP coefficients using the equation:
970 * y[i] = x[i] * (gamma ** i)
972 * @param[out] out Output buffer (may use input array)
973 * @param[in] lpc LP coefficients array
974 * @param[in] gamma Weighting factor
975 * @param[in] size LP array size
977 static void lpc_weighting(float *out, const float *lpc, float gamma, int size)
982 for (i = 0; i < size; i++) {
983 out[i] = lpc[i] * fac;
989 * Conduct 20th order linear predictive coding synthesis for the high
990 * frequency band excitation at 16kHz.
992 * @param[in] ctx The context
993 * @param[in] subframe Current subframe index (0 to 3)
994 * @param[in,out] samples Pointer to the output speech samples
995 * @param[in] exc Generated white-noise scaled excitation
996 * @param[in] isf Current frame isf vector
997 * @param[in] isf_past Past frame final isf vector
999 static void hb_synthesis(AMRWBContext *ctx, int subframe, float *samples,
1000 const float *exc, const float *isf, const float *isf_past)
1002 float hb_lpc[LP_ORDER_16k];
1003 enum Mode mode = ctx->fr_cur_mode;
1005 if (mode == MODE_6k60) {
1006 float e_isf[LP_ORDER_16k]; // ISF vector for extrapolation
1007 double e_isp[LP_ORDER_16k];
1009 ff_weighted_vector_sumf(e_isf, isf_past, isf, isfp_inter[subframe],
1010 1.0 - isfp_inter[subframe], LP_ORDER);
1012 extrapolate_isf(e_isf);
1014 e_isf[LP_ORDER_16k - 1] *= 2.0;
1015 ff_acelp_lsf2lspd(e_isp, e_isf, LP_ORDER_16k);
1016 ff_amrwb_lsp2lpc(e_isp, hb_lpc, LP_ORDER_16k);
1018 lpc_weighting(hb_lpc, hb_lpc, 0.9, LP_ORDER_16k);
1020 lpc_weighting(hb_lpc, ctx->lp_coef[subframe], 0.6, LP_ORDER);
1023 ff_celp_lp_synthesis_filterf(samples, hb_lpc, exc, AMRWB_SFR_SIZE_16k,
1024 (mode == MODE_6k60) ? LP_ORDER_16k : LP_ORDER);
1028 * Apply a 15th order filter to high-band samples.
1029 * The filter characteristic depends on the given coefficients.
1031 * @param[out] out Buffer for filtered output
1032 * @param[in] fir_coef Filter coefficients
1033 * @param[in,out] mem State from last filtering (updated)
1034 * @param[in] in Input speech data (high-band)
1036 * @remark It is safe to pass the same array in in and out parameters
1038 static void hb_fir_filter(float *out, const float fir_coef[HB_FIR_SIZE + 1],
1039 float mem[HB_FIR_SIZE], const float *in)
1042 float data[AMRWB_SFR_SIZE_16k + HB_FIR_SIZE]; // past and current samples
1044 memcpy(data, mem, HB_FIR_SIZE * sizeof(float));
1045 memcpy(data + HB_FIR_SIZE, in, AMRWB_SFR_SIZE_16k * sizeof(float));
1047 for (i = 0; i < AMRWB_SFR_SIZE_16k; i++) {
1049 for (j = 0; j <= HB_FIR_SIZE; j++)
1050 out[i] += data[i + j] * fir_coef[j];
1053 memcpy(mem, data + AMRWB_SFR_SIZE_16k, HB_FIR_SIZE * sizeof(float));
1057 * Update context state before the next subframe.
1059 static void update_sub_state(AMRWBContext *ctx)
1061 memmove(&ctx->excitation_buf[0], &ctx->excitation_buf[AMRWB_SFR_SIZE],
1062 (AMRWB_P_DELAY_MAX + LP_ORDER + 1) * sizeof(float));
1064 memmove(&ctx->pitch_gain[1], &ctx->pitch_gain[0], 5 * sizeof(float));
1065 memmove(&ctx->fixed_gain[1], &ctx->fixed_gain[0], 1 * sizeof(float));
1067 memmove(&ctx->samples_az[0], &ctx->samples_az[AMRWB_SFR_SIZE],
1068 LP_ORDER * sizeof(float));
1069 memmove(&ctx->samples_up[0], &ctx->samples_up[AMRWB_SFR_SIZE],
1070 UPS_MEM_SIZE * sizeof(float));
1071 memmove(&ctx->samples_hb[0], &ctx->samples_hb[AMRWB_SFR_SIZE_16k],
1072 LP_ORDER_16k * sizeof(float));
1075 static int amrwb_decode_frame(AVCodecContext *avctx, void *data,
1076 int *got_frame_ptr, AVPacket *avpkt)
1078 AMRWBContext *ctx = avctx->priv_data;
1079 AMRWBFrame *cf = &ctx->frame;
1080 const uint8_t *buf = avpkt->data;
1081 int buf_size = avpkt->size;
1082 int expected_fr_size, header_size;
1084 float spare_vector[AMRWB_SFR_SIZE]; // extra stack space to hold result from anti-sparseness processing
1085 float fixed_gain_factor; // fixed gain correction factor (gamma)
1086 float *synth_fixed_vector; // pointer to the fixed vector that synthesis should use
1087 float synth_fixed_gain; // the fixed gain that synthesis should use
1088 float voice_fac, stab_fac; // parameters used for gain smoothing
1089 float synth_exc[AMRWB_SFR_SIZE]; // post-processed excitation for synthesis
1090 float hb_exc[AMRWB_SFR_SIZE_16k]; // excitation for the high frequency band
1091 float hb_samples[AMRWB_SFR_SIZE_16k]; // filtered high-band samples from synthesis
1095 /* get output buffer */
1096 ctx->avframe.nb_samples = 4 * AMRWB_SFR_SIZE_16k;
1097 if ((ret = ff_get_buffer(avctx, &ctx->avframe)) < 0) {
1098 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1101 buf_out = (float *)ctx->avframe.data[0];
1103 header_size = decode_mime_header(ctx, buf);
1104 if (ctx->fr_cur_mode > MODE_SID) {
1105 av_log(avctx, AV_LOG_ERROR,
1106 "Invalid mode %d\n", ctx->fr_cur_mode);
1107 return AVERROR_INVALIDDATA;
1109 expected_fr_size = ((cf_sizes_wb[ctx->fr_cur_mode] + 7) >> 3) + 1;
1111 if (buf_size < expected_fr_size) {
1112 av_log(avctx, AV_LOG_ERROR,
1113 "Frame too small (%d bytes). Truncated file?\n", buf_size);
1115 return AVERROR_INVALIDDATA;
1118 if (!ctx->fr_quality || ctx->fr_cur_mode > MODE_SID)
1119 av_log(avctx, AV_LOG_ERROR, "Encountered a bad or corrupted frame\n");
1121 if (ctx->fr_cur_mode == MODE_SID) { /* Comfort noise frame */
1122 av_log_missing_feature(avctx, "SID mode", 1);
1123 return AVERROR_PATCHWELCOME;
1126 ff_amr_bit_reorder((uint16_t *) &ctx->frame, sizeof(AMRWBFrame),
1127 buf + header_size, amr_bit_orderings_by_mode[ctx->fr_cur_mode]);
1129 /* Decode the quantized ISF vector */
1130 if (ctx->fr_cur_mode == MODE_6k60) {
1131 decode_isf_indices_36b(cf->isp_id, ctx->isf_cur);
1133 decode_isf_indices_46b(cf->isp_id, ctx->isf_cur);
1136 isf_add_mean_and_past(ctx->isf_cur, ctx->isf_q_past);
1137 ff_set_min_dist_lsf(ctx->isf_cur, MIN_ISF_SPACING, LP_ORDER - 1);
1139 stab_fac = stability_factor(ctx->isf_cur, ctx->isf_past_final);
1141 ctx->isf_cur[LP_ORDER - 1] *= 2.0;
1142 ff_acelp_lsf2lspd(ctx->isp[3], ctx->isf_cur, LP_ORDER);
1144 /* Generate a ISP vector for each subframe */
1145 if (ctx->first_frame) {
1146 ctx->first_frame = 0;
1147 memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(double));
1149 interpolate_isp(ctx->isp, ctx->isp_sub4_past);
1151 for (sub = 0; sub < 4; sub++)
1152 ff_amrwb_lsp2lpc(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER);
1154 for (sub = 0; sub < 4; sub++) {
1155 const AMRWBSubFrame *cur_subframe = &cf->subframe[sub];
1156 float *sub_buf = buf_out + sub * AMRWB_SFR_SIZE_16k;
1158 /* Decode adaptive codebook (pitch vector) */
1159 decode_pitch_vector(ctx, cur_subframe, sub);
1160 /* Decode innovative codebook (fixed vector) */
1161 decode_fixed_vector(ctx->fixed_vector, cur_subframe->pul_ih,
1162 cur_subframe->pul_il, ctx->fr_cur_mode);
1164 pitch_sharpening(ctx, ctx->fixed_vector);
1166 decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode,
1167 &fixed_gain_factor, &ctx->pitch_gain[0]);
1169 ctx->fixed_gain[0] =
1170 ff_amr_set_fixed_gain(fixed_gain_factor,
1171 ff_scalarproduct_float_c(ctx->fixed_vector,
1175 ctx->prediction_error,
1176 ENERGY_MEAN, energy_pred_fac);
1178 /* Calculate voice factor and store tilt for next subframe */
1179 voice_fac = voice_factor(ctx->pitch_vector, ctx->pitch_gain[0],
1180 ctx->fixed_vector, ctx->fixed_gain[0]);
1181 ctx->tilt_coef = voice_fac * 0.25 + 0.25;
1183 /* Construct current excitation */
1184 for (i = 0; i < AMRWB_SFR_SIZE; i++) {
1185 ctx->excitation[i] *= ctx->pitch_gain[0];
1186 ctx->excitation[i] += ctx->fixed_gain[0] * ctx->fixed_vector[i];
1187 ctx->excitation[i] = truncf(ctx->excitation[i]);
1190 /* Post-processing of excitation elements */
1191 synth_fixed_gain = noise_enhancer(ctx->fixed_gain[0], &ctx->prev_tr_gain,
1192 voice_fac, stab_fac);
1194 synth_fixed_vector = anti_sparseness(ctx, ctx->fixed_vector,
1197 pitch_enhancer(synth_fixed_vector, voice_fac);
1199 synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain,
1200 synth_fixed_vector, &ctx->samples_az[LP_ORDER]);
1202 /* Synthesis speech post-processing */
1203 de_emphasis(&ctx->samples_up[UPS_MEM_SIZE],
1204 &ctx->samples_az[LP_ORDER], PREEMPH_FAC, ctx->demph_mem);
1206 ff_acelp_apply_order_2_transfer_function(&ctx->samples_up[UPS_MEM_SIZE],
1207 &ctx->samples_up[UPS_MEM_SIZE], hpf_zeros, hpf_31_poles,
1208 hpf_31_gain, ctx->hpf_31_mem, AMRWB_SFR_SIZE);
1210 upsample_5_4(sub_buf, &ctx->samples_up[UPS_FIR_SIZE],
1211 AMRWB_SFR_SIZE_16k);
1213 /* High frequency band (6.4 - 7.0 kHz) generation part */
1214 ff_acelp_apply_order_2_transfer_function(hb_samples,
1215 &ctx->samples_up[UPS_MEM_SIZE], hpf_zeros, hpf_400_poles,
1216 hpf_400_gain, ctx->hpf_400_mem, AMRWB_SFR_SIZE);
1218 hb_gain = find_hb_gain(ctx, hb_samples,
1219 cur_subframe->hb_gain, cf->vad);
1221 scaled_hb_excitation(ctx, hb_exc, synth_exc, hb_gain);
1223 hb_synthesis(ctx, sub, &ctx->samples_hb[LP_ORDER_16k],
1224 hb_exc, ctx->isf_cur, ctx->isf_past_final);
1226 /* High-band post-processing filters */
1227 hb_fir_filter(hb_samples, bpf_6_7_coef, ctx->bpf_6_7_mem,
1228 &ctx->samples_hb[LP_ORDER_16k]);
1230 if (ctx->fr_cur_mode == MODE_23k85)
1231 hb_fir_filter(hb_samples, lpf_7_coef, ctx->lpf_7_mem,
1234 /* Add the low and high frequency bands */
1235 for (i = 0; i < AMRWB_SFR_SIZE_16k; i++)
1236 sub_buf[i] = (sub_buf[i] + hb_samples[i]) * (1.0f / (1 << 15));
1238 /* Update buffers and history */
1239 update_sub_state(ctx);
1242 /* update state for next frame */
1243 memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(ctx->isp[3][0]));
1244 memcpy(ctx->isf_past_final, ctx->isf_cur, LP_ORDER * sizeof(float));
1247 *(AVFrame *)data = ctx->avframe;
1249 return expected_fr_size;
1252 AVCodec ff_amrwb_decoder = {
1254 .type = AVMEDIA_TYPE_AUDIO,
1255 .id = AV_CODEC_ID_AMR_WB,
1256 .priv_data_size = sizeof(AMRWBContext),
1257 .init = amrwb_decode_init,
1258 .decode = amrwb_decode_frame,
1259 .capabilities = CODEC_CAP_DR1,
1260 .long_name = NULL_IF_CONFIG_SMALL("AMR-WB (Adaptive Multi-Rate WideBand)"),
1261 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLT,
1262 AV_SAMPLE_FMT_NONE },