Mercurial > libavcodec.hg
comparison sipr.c @ 10836:f20f9b76f4c8 libavcodec
SIPR decoder for modes 8k5, 6k5 and 5k0.
Patch by Vladimir Voroshilov and myself.
author | vitor |
---|---|
date | Sun, 10 Jan 2010 21:38:15 +0000 |
parents | |
children | c06d95ce2db0 |
comparison
equal
deleted
inserted
replaced
10835:d1115fb7e914 | 10836:f20f9b76f4c8 |
---|---|
1 /* | |
2 * SIPR / ACELP.NET decoder | |
3 * | |
4 * Copyright (c) 2008 Vladimir Voroshilov | |
5 * Copyright (c) 2009 Vitor Sessak | |
6 * | |
7 * This file is part of FFmpeg. | |
8 * | |
9 * FFmpeg is free software; you can redistribute it and/or | |
10 * modify it under the terms of the GNU Lesser General Public | |
11 * License as published by the Free Software Foundation; either | |
12 * version 2.1 of the License, or (at your option) any later version. | |
13 * | |
14 * FFmpeg is distributed in the hope that it will be useful, | |
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 * Lesser General Public License for more details. | |
18 * | |
19 * You should have received a copy of the GNU Lesser General Public | |
20 * License along with FFmpeg; if not, write to the Free Software | |
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 */ | |
23 | |
24 #include <math.h> | |
25 #include <stdint.h> | |
26 | |
27 #include "avcodec.h" | |
28 #define ALT_BITSTREAM_READER_LE | |
29 #include "get_bits.h" | |
30 #include "dsputil.h" | |
31 | |
32 #include "lsp.h" | |
33 #include "celp_math.h" | |
34 #include "acelp_vectors.h" | |
35 #include "acelp_pitch_delay.h" | |
36 #include "acelp_filters.h" | |
37 #include "celp_filters.h" | |
38 | |
39 #define LSFQ_DIFF_MIN (0.0125 * M_PI) | |
40 | |
41 #define LP_FILTER_ORDER 10 | |
42 | |
43 /** Number of past samples needed for excitation interpolation */ | |
44 #define L_INTERPOL (LP_FILTER_ORDER + 1) | |
45 | |
46 /** Subframe size for all modes except 16k */ | |
47 #define SUBFR_SIZE 48 | |
48 | |
49 #include "siprdata.h" | |
50 | |
51 typedef enum { | |
52 MODE_16k, | |
53 MODE_8k5, | |
54 MODE_6k5, | |
55 MODE_5k0, | |
56 MODE_COUNT | |
57 } SiprMode; | |
58 | |
59 typedef struct { | |
60 const char *mode_name; | |
61 uint16_t bits_per_frame; | |
62 uint8_t subframe_count; | |
63 uint8_t frames_per_packet; | |
64 float pitch_sharp_factor; | |
65 | |
66 /* bitstream parameters */ | |
67 uint8_t number_of_fc_indexes; | |
68 | |
69 /** size in bits of the i-th stage vector of quantizer */ | |
70 uint8_t vq_indexes_bits[5]; | |
71 | |
72 /** size in bits of the adaptive-codebook index for every subframe */ | |
73 uint8_t pitch_delay_bits[5]; | |
74 | |
75 uint8_t gp_index_bits; | |
76 uint8_t fc_index_bits[10]; ///< size in bits of the fixed codebook indexes | |
77 uint8_t gc_index_bits; ///< size in bits of the gain codebook indexes | |
78 } SiprModeParam; | |
79 | |
80 static const SiprModeParam modes[MODE_COUNT] = { | |
81 [MODE_8k5] = { | |
82 .mode_name = "8k5", | |
83 .bits_per_frame = 152, | |
84 .subframe_count = 3, | |
85 .frames_per_packet = 1, | |
86 .pitch_sharp_factor = 0.8, | |
87 | |
88 .number_of_fc_indexes = 3, | |
89 .vq_indexes_bits = {6, 7, 7, 7, 5}, | |
90 .pitch_delay_bits = {8, 5, 5}, | |
91 .gp_index_bits = 0, | |
92 .fc_index_bits = {9, 9, 9}, | |
93 .gc_index_bits = 7 | |
94 }, | |
95 | |
96 [MODE_6k5] = { | |
97 .mode_name = "6k5", | |
98 .bits_per_frame = 232, | |
99 .subframe_count = 3, | |
100 .frames_per_packet = 2, | |
101 .pitch_sharp_factor = 0.8, | |
102 | |
103 .number_of_fc_indexes = 3, | |
104 .vq_indexes_bits = {6, 7, 7, 7, 5}, | |
105 .pitch_delay_bits = {8, 5, 5}, | |
106 .gp_index_bits = 0, | |
107 .fc_index_bits = {5, 5, 5}, | |
108 .gc_index_bits = 7 | |
109 }, | |
110 | |
111 [MODE_5k0] = { | |
112 .mode_name = "5k0", | |
113 .bits_per_frame = 296, | |
114 .subframe_count = 5, | |
115 .frames_per_packet = 2, | |
116 .pitch_sharp_factor = 0.85, | |
117 | |
118 .number_of_fc_indexes = 1, | |
119 .vq_indexes_bits = {6, 7, 7, 7, 5}, | |
120 .pitch_delay_bits = {8, 5, 8, 5, 5}, | |
121 .gp_index_bits = 0, | |
122 .fc_index_bits = {10}, | |
123 .gc_index_bits = 7 | |
124 } | |
125 }; | |
126 | |
127 typedef struct { | |
128 AVCodecContext *avctx; | |
129 DSPContext dsp; | |
130 | |
131 SiprModeParam m; | |
132 SiprMode mode; | |
133 | |
134 float past_pitch_gain; | |
135 float lsf_history[LP_FILTER_ORDER]; | |
136 | |
137 float excitation[L_INTERPOL + PITCH_DELAY_MAX + 5*SUBFR_SIZE]; | |
138 | |
139 DECLARE_ALIGNED_16(float, synth_buf[LP_FILTER_ORDER + 5*SUBFR_SIZE + 6]); | |
140 | |
141 float lsp_history[LP_FILTER_ORDER]; | |
142 float gain_mem; | |
143 float energy_history[4]; | |
144 float highpass_filt_mem[2]; | |
145 float postfilter_mem[PITCH_DELAY_MAX + LP_FILTER_ORDER]; | |
146 | |
147 /* 5k0 */ | |
148 float tilt_mem; | |
149 float postfilter_agc; | |
150 float postfilter_mem5k0[PITCH_DELAY_MAX + LP_FILTER_ORDER]; | |
151 float postfilter_syn5k0[LP_FILTER_ORDER + SUBFR_SIZE*5]; | |
152 } SiprContext; | |
153 | |
154 typedef struct { | |
155 int vq_indexes[5]; | |
156 int pitch_delay[5]; ///< pitch delay | |
157 int gp_index[5]; ///< adaptive-codebook gain indexes | |
158 int16_t fc_indexes[5][10]; ///< fixed-codebook indexes | |
159 int gc_index[5]; ///< fixed-codebook gain indexes | |
160 } SiprParameters; | |
161 | |
162 | |
163 static void dequant(float *out, const int *idx, const float *cbs[]) | |
164 { | |
165 int i; | |
166 int stride = 2; | |
167 int num_vec = 5; | |
168 | |
169 for (i = 0; i < num_vec; i++) | |
170 memcpy(out + stride*i, cbs[i] + stride*idx[i], stride*sizeof(float)); | |
171 | |
172 } | |
173 | |
174 static void lsf_decode_fp(float *lsfnew, float *lsf_history, | |
175 const SiprParameters *parm) | |
176 { | |
177 int i; | |
178 float lsf_tmp[LP_FILTER_ORDER]; | |
179 | |
180 dequant(lsf_tmp, parm->vq_indexes, lsf_codebooks); | |
181 | |
182 for (i = 0; i < LP_FILTER_ORDER; i++) | |
183 lsfnew[i] = lsf_history[i] * 0.33 + lsf_tmp[i] + mean_lsf[i]; | |
184 | |
185 ff_sort_nearly_sorted_floats(lsfnew, LP_FILTER_ORDER - 1); | |
186 | |
187 /* Note that a minimum distance is not enforced between the last value and | |
188 the previous one, contrary to what is done in ff_acelp_reorder_lsf() */ | |
189 ff_set_min_dist_lsf(lsfnew, LSFQ_DIFF_MIN, LP_FILTER_ORDER - 1); | |
190 lsfnew[9] = FFMIN(lsfnew[LP_FILTER_ORDER - 1], 1.3 * M_PI); | |
191 | |
192 memcpy(lsf_history, lsf_tmp, LP_FILTER_ORDER * sizeof(*lsf_history)); | |
193 | |
194 for (i = 0; i < LP_FILTER_ORDER - 1; i++) | |
195 lsfnew[i] = cos(lsfnew[i]); | |
196 lsfnew[LP_FILTER_ORDER - 1] *= 6.153848 / M_PI; | |
197 } | |
198 | |
199 /** Apply pitch lag to the fixed vector (AMR section 6.1.2). */ | |
200 static void pitch_sharpening(int pitch_lag_int, float beta, | |
201 float *fixed_vector) | |
202 { | |
203 int i; | |
204 | |
205 for (i = pitch_lag_int; i < SUBFR_SIZE; i++) | |
206 fixed_vector[i] += beta * fixed_vector[i - pitch_lag_int]; | |
207 } | |
208 | |
209 /** | |
210 * Extracts decoding parameters from the input bitstream. | |
211 * @param parms parameters structure | |
212 * @param pgb pointer to initialized GetBitContext structure | |
213 */ | |
214 static void decode_parameters(SiprParameters* parms, GetBitContext *pgb, | |
215 const SiprModeParam *p) | |
216 { | |
217 int i, j; | |
218 | |
219 for (i = 0; i < 5; i++) | |
220 parms->vq_indexes[i] = get_bits(pgb, p->vq_indexes_bits[i]); | |
221 | |
222 for (i = 0; i < p->subframe_count; i++) { | |
223 parms->pitch_delay[i] = get_bits(pgb, p->pitch_delay_bits[i]); | |
224 parms->gp_index[i] = get_bits(pgb, p->gp_index_bits); | |
225 | |
226 for (j = 0; j < p->number_of_fc_indexes; j++) | |
227 parms->fc_indexes[i][j] = get_bits(pgb, p->fc_index_bits[j]); | |
228 | |
229 parms->gc_index[i] = get_bits(pgb, p->gc_index_bits); | |
230 } | |
231 } | |
232 | |
233 static void lsp2lpc_sipr(const double *lsp, float *Az) | |
234 { | |
235 int lp_half_order = LP_FILTER_ORDER >> 1; | |
236 double buf[lp_half_order + 1]; | |
237 double pa[lp_half_order + 1]; | |
238 double *qa = buf + 1; | |
239 int i,j; | |
240 | |
241 qa[-1] = 0.0; | |
242 | |
243 ff_lsp2polyf(lsp , pa, lp_half_order ); | |
244 ff_lsp2polyf(lsp + 1, qa, lp_half_order - 1); | |
245 | |
246 for (i = 1, j = LP_FILTER_ORDER - 1; i < lp_half_order; i++, j--) { | |
247 double paf = pa[i] * (1 + lsp[LP_FILTER_ORDER - 1]); | |
248 double qaf = (qa[i] - qa[i-2]) * (1 - lsp[LP_FILTER_ORDER - 1]); | |
249 Az[i-1] = (paf + qaf) * 0.5; | |
250 Az[j-1] = (paf - qaf) * 0.5; | |
251 } | |
252 | |
253 Az[lp_half_order - 1] = (1.0 + lsp[LP_FILTER_ORDER - 1]) * | |
254 pa[lp_half_order] * 0.5; | |
255 | |
256 Az[LP_FILTER_ORDER - 1] = lsp[LP_FILTER_ORDER - 1]; | |
257 } | |
258 | |
259 static void sipr_decode_lp(float *lsfnew, const float *lsfold, float *Az, | |
260 int num_subfr) | |
261 { | |
262 double lsfint[LP_FILTER_ORDER]; | |
263 int i,j; | |
264 float t, t0 = 1.0 / num_subfr; | |
265 | |
266 t = t0 * 0.5; | |
267 for (i = 0; i < num_subfr; i++) { | |
268 for (j = 0; j < LP_FILTER_ORDER; j++) | |
269 lsfint[j] = lsfold[j] * (1 - t) + t * lsfnew[j]; | |
270 | |
271 lsp2lpc_sipr(lsfint, Az); | |
272 Az += LP_FILTER_ORDER; | |
273 t += t0; | |
274 } | |
275 } | |
276 | |
277 /** | |
278 * Evaluates the adaptative impulse response. | |
279 */ | |
280 static void eval_ir(const float *Az, int pitch_lag, float *freq, | |
281 float pitch_sharp_factor) | |
282 { | |
283 float tmp1[SUBFR_SIZE+1], tmp2[LP_FILTER_ORDER+1]; | |
284 int i; | |
285 | |
286 tmp1[0] = 1.; | |
287 for (i = 0; i < LP_FILTER_ORDER; i++) { | |
288 tmp1[i+1] = Az[i] * ff_pow_0_55[i]; | |
289 tmp2[i ] = Az[i] * ff_pow_0_7 [i]; | |
290 } | |
291 memset(tmp1 + 11, 0, 37 * sizeof(float)); | |
292 | |
293 ff_celp_lp_synthesis_filterf(freq, tmp2, tmp1, SUBFR_SIZE, | |
294 LP_FILTER_ORDER); | |
295 | |
296 pitch_sharpening(pitch_lag, pitch_sharp_factor, freq); | |
297 } | |
298 | |
299 /** | |
300 * Evaluates the convolution of a vector with a sparse vector. | |
301 */ | |
302 static void convolute_with_sparse(float *out, const AMRFixed *pulses, | |
303 const float *shape, int length) | |
304 { | |
305 int i, j; | |
306 | |
307 memset(out, 0, length*sizeof(float)); | |
308 for (i = 0; i < pulses->n; i++) | |
309 for (j = pulses->x[i]; j < length; j++) | |
310 out[j] += pulses->y[i] * shape[j - pulses->x[i]]; | |
311 } | |
312 | |
313 /** | |
314 * Apply postfilter, very similar to AMR one. | |
315 */ | |
316 static void postfilter_5k0(SiprContext *ctx, const float *lpc, float *samples) | |
317 { | |
318 float buf[SUBFR_SIZE + LP_FILTER_ORDER]; | |
319 float *pole_out = buf + LP_FILTER_ORDER; | |
320 float lpc_n[LP_FILTER_ORDER]; | |
321 float lpc_d[LP_FILTER_ORDER]; | |
322 int i; | |
323 | |
324 for (i = 0; i < LP_FILTER_ORDER; i++) { | |
325 lpc_d[i] = lpc[i] * ff_pow_0_75[i]; | |
326 lpc_n[i] = lpc[i] * pow_0_5 [i]; | |
327 }; | |
328 | |
329 memcpy(pole_out - LP_FILTER_ORDER, ctx->postfilter_mem, | |
330 LP_FILTER_ORDER*sizeof(float)); | |
331 | |
332 ff_celp_lp_synthesis_filterf(pole_out, lpc_d, samples, SUBFR_SIZE, | |
333 LP_FILTER_ORDER); | |
334 | |
335 memcpy(ctx->postfilter_mem, pole_out + SUBFR_SIZE - LP_FILTER_ORDER, | |
336 LP_FILTER_ORDER*sizeof(float)); | |
337 | |
338 ff_tilt_compensation(&ctx->tilt_mem, 0.4, pole_out, SUBFR_SIZE); | |
339 | |
340 memcpy(pole_out - LP_FILTER_ORDER, ctx->postfilter_mem5k0, | |
341 LP_FILTER_ORDER*sizeof(*pole_out)); | |
342 | |
343 memcpy(ctx->postfilter_mem5k0, pole_out + SUBFR_SIZE - LP_FILTER_ORDER, | |
344 LP_FILTER_ORDER*sizeof(*pole_out)); | |
345 | |
346 ff_celp_lp_zero_synthesis_filterf(samples, lpc_n, pole_out, SUBFR_SIZE, | |
347 LP_FILTER_ORDER); | |
348 | |
349 } | |
350 | |
351 static void decode_fixed_sparse(AMRFixed *fixed_sparse, const int16_t *pulses, | |
352 SiprMode mode, int low_gain) | |
353 { | |
354 int i; | |
355 | |
356 switch (mode) { | |
357 case MODE_6k5: | |
358 for (i = 0; i < 3; i++) { | |
359 fixed_sparse->x[i] = 3 * (pulses[i] & 0xf) + i; | |
360 fixed_sparse->y[i] = pulses[i] & 0x10 ? -1 : 1; | |
361 } | |
362 fixed_sparse->n = 3; | |
363 break; | |
364 case MODE_8k5: | |
365 for (i = 0; i < 3; i++) { | |
366 fixed_sparse->x[2*i ] = 3 * ((pulses[i] >> 4) & 0xf) + i; | |
367 fixed_sparse->x[2*i + 1] = 3 * ( pulses[i] & 0xf) + i; | |
368 | |
369 fixed_sparse->y[2*i ] = (pulses[i] & 0x100) ? -1.0: 1.0; | |
370 | |
371 fixed_sparse->y[2*i + 1] = | |
372 (fixed_sparse->x[2*i + 1] < fixed_sparse->x[2*i]) ? | |
373 -fixed_sparse->y[2*i ] : fixed_sparse->y[2*i]; | |
374 } | |
375 | |
376 fixed_sparse->n = 6; | |
377 break; | |
378 case MODE_5k0: | |
379 default: | |
380 if (low_gain) { | |
381 int offset = (pulses[0] & 0x200) ? 2 : 0; | |
382 int val = pulses[0]; | |
383 | |
384 for (i = 0; i < 3; i++) { | |
385 int index = (val & 0x7) * 6 + 4 - i*2; | |
386 | |
387 fixed_sparse->y[i] = (offset + index) & 0x3 ? -1 : 1; | |
388 fixed_sparse->x[i] = index; | |
389 | |
390 val >>= 3; | |
391 } | |
392 fixed_sparse->n = 3; | |
393 } else { | |
394 int pulse_subset = (pulses[0] >> 8) & 1; | |
395 | |
396 fixed_sparse->x[0] = ((pulses[0] >> 4) & 15) * 3 + pulse_subset; | |
397 fixed_sparse->x[1] = ( pulses[0] & 15) * 3 + pulse_subset + 1; | |
398 | |
399 fixed_sparse->y[0] = pulses[0] & 0x200 ? -1 : 1; | |
400 fixed_sparse->y[1] = -fixed_sparse->y[0]; | |
401 fixed_sparse->n = 2; | |
402 } | |
403 break; | |
404 } | |
405 } | |
406 | |
407 static void decode_frame(SiprContext *ctx, SiprParameters *params, | |
408 float *out_data) | |
409 { | |
410 int i, j; | |
411 int frame_size = ctx->m.subframe_count * SUBFR_SIZE; | |
412 float Az[LP_FILTER_ORDER * ctx->m.subframe_count]; | |
413 float *excitation; | |
414 float ir_buf[SUBFR_SIZE + LP_FILTER_ORDER]; | |
415 float lsf_new[LP_FILTER_ORDER]; | |
416 float *impulse_response = ir_buf + LP_FILTER_ORDER; | |
417 float *synth = ctx->synth_buf + 16; // 16 instead of LP_FILTER_ORDER for | |
418 // memory alignment | |
419 int t0_first = 0; | |
420 AMRFixed fixed_cb; | |
421 | |
422 memset(ir_buf, 0, LP_FILTER_ORDER * sizeof(float)); | |
423 lsf_decode_fp(lsf_new, ctx->lsf_history, params); | |
424 | |
425 sipr_decode_lp(lsf_new, ctx->lsp_history, Az, ctx->m.subframe_count); | |
426 | |
427 memcpy(ctx->lsp_history, lsf_new, LP_FILTER_ORDER * sizeof(float)); | |
428 | |
429 excitation = ctx->excitation + PITCH_DELAY_MAX + L_INTERPOL; | |
430 | |
431 for (i = 0; i < ctx->m.subframe_count; i++) { | |
432 float *pAz = Az + i*LP_FILTER_ORDER; | |
433 float fixed_vector[SUBFR_SIZE]; | |
434 int T0,T0_frac; | |
435 float pitch_gain, gain_code, avg_energy; | |
436 | |
437 ff_decode_pitch_lag(&T0, &T0_frac, params->pitch_delay[i], t0_first, i, | |
438 ctx->mode == MODE_5k0, 6); | |
439 | |
440 if (i == 0 || (i == 2 && ctx->mode == MODE_5k0)) | |
441 t0_first = T0; | |
442 | |
443 ff_acelp_interpolatef(excitation, excitation - T0 + (T0_frac <= 0), | |
444 ff_b60_sinc, 6, | |
445 2 * ((2 + T0_frac)%3 + 1), LP_FILTER_ORDER, | |
446 SUBFR_SIZE); | |
447 | |
448 decode_fixed_sparse(&fixed_cb, params->fc_indexes[i], ctx->mode, | |
449 ctx->past_pitch_gain < 0.8); | |
450 | |
451 eval_ir(pAz, T0, impulse_response, ctx->m.pitch_sharp_factor); | |
452 | |
453 convolute_with_sparse(fixed_vector, &fixed_cb, impulse_response, | |
454 SUBFR_SIZE); | |
455 | |
456 avg_energy = | |
457 (0.01 + ff_dot_productf(fixed_vector, fixed_vector, SUBFR_SIZE))/ | |
458 SUBFR_SIZE; | |
459 | |
460 ctx->past_pitch_gain = pitch_gain = gain_cb[params->gc_index[i]][0]; | |
461 | |
462 gain_code = ff_amr_set_fixed_gain(gain_cb[params->gc_index[i]][1], | |
463 avg_energy, ctx->energy_history, | |
464 34 - 15.0/(log2f(10.0) * 0.05), | |
465 pred); | |
466 | |
467 ff_weighted_vector_sumf(excitation, excitation, fixed_vector, | |
468 pitch_gain, gain_code, SUBFR_SIZE); | |
469 | |
470 pitch_gain *= 0.5 * pitch_gain; | |
471 pitch_gain = FFMIN(pitch_gain, 0.4); | |
472 | |
473 ctx->gain_mem = 0.7 * ctx->gain_mem + 0.3 * pitch_gain; | |
474 ctx->gain_mem = FFMIN(ctx->gain_mem, pitch_gain); | |
475 gain_code *= ctx->gain_mem; | |
476 | |
477 for (j = 0; j < SUBFR_SIZE; j++) | |
478 fixed_vector[j] = excitation[j] - gain_code * fixed_vector[j]; | |
479 | |
480 if (ctx->mode == MODE_5k0) { | |
481 postfilter_5k0(ctx, pAz, fixed_vector); | |
482 | |
483 ff_celp_lp_synthesis_filterf(ctx->postfilter_syn5k0 + LP_FILTER_ORDER + i*SUBFR_SIZE, | |
484 pAz, excitation, SUBFR_SIZE, | |
485 LP_FILTER_ORDER); | |
486 } | |
487 | |
488 ff_celp_lp_synthesis_filterf(synth + i*SUBFR_SIZE, pAz, fixed_vector, | |
489 SUBFR_SIZE, LP_FILTER_ORDER); | |
490 | |
491 excitation += SUBFR_SIZE; | |
492 } | |
493 | |
494 memcpy(synth - LP_FILTER_ORDER, synth + frame_size - LP_FILTER_ORDER, | |
495 LP_FILTER_ORDER * sizeof(float)); | |
496 | |
497 if (ctx->mode == MODE_5k0) { | |
498 for (i = 0; i < ctx->m.subframe_count; i++) { | |
499 float energy = ff_dot_productf(ctx->postfilter_syn5k0 + LP_FILTER_ORDER + i*SUBFR_SIZE, | |
500 ctx->postfilter_syn5k0 + LP_FILTER_ORDER + i*SUBFR_SIZE, | |
501 SUBFR_SIZE); | |
502 ff_adaptative_gain_control(&synth[i * SUBFR_SIZE], energy, | |
503 SUBFR_SIZE, 0.9, &ctx->postfilter_agc); | |
504 } | |
505 | |
506 memcpy(ctx->postfilter_syn5k0, ctx->postfilter_syn5k0 + frame_size, | |
507 LP_FILTER_ORDER*sizeof(float)); | |
508 } | |
509 memcpy(ctx->excitation, excitation - PITCH_DELAY_MAX - L_INTERPOL, | |
510 (PITCH_DELAY_MAX + L_INTERPOL) * sizeof(float)); | |
511 | |
512 ff_acelp_apply_order_2_transfer_function(synth, | |
513 (const float[2]) {-1.99997 , 1.000000000}, | |
514 (const float[2]) {-1.93307352, 0.935891986}, | |
515 0.939805806, | |
516 ctx->highpass_filt_mem, | |
517 frame_size); | |
518 | |
519 ctx->dsp.vector_clipf(out_data, synth, -1, 32767./(1<<15), frame_size); | |
520 | |
521 } | |
522 | |
523 static av_cold int sipr_decoder_init(AVCodecContext * avctx) | |
524 { | |
525 SiprContext *ctx = avctx->priv_data; | |
526 int i; | |
527 | |
528 if (avctx->bit_rate > 12200) ctx->mode = MODE_16k; | |
529 else if (avctx->bit_rate > 7500 ) ctx->mode = MODE_8k5; | |
530 else if (avctx->bit_rate > 5750 ) ctx->mode = MODE_6k5; | |
531 else ctx->mode = MODE_5k0; | |
532 | |
533 ctx->m = modes[ctx->mode]; | |
534 av_log(avctx, AV_LOG_DEBUG, "Mode: %s\n", ctx->m.mode_name); | |
535 | |
536 for (i = 0; i < LP_FILTER_ORDER; i++) | |
537 ctx->lsp_history[i] = cos((i+1) * M_PI / (LP_FILTER_ORDER + 1)); | |
538 | |
539 for (i = 0; i < 4; i++) | |
540 ctx->energy_history[i] = -14; | |
541 | |
542 avctx->sample_fmt = SAMPLE_FMT_FLT; | |
543 | |
544 if (ctx->mode == MODE_16k) { | |
545 av_log(avctx, AV_LOG_ERROR, "decoding 16kbps SIPR files is not " | |
546 "supported yet.\n"); | |
547 return -1; | |
548 } | |
549 | |
550 dsputil_init(&ctx->dsp, avctx); | |
551 | |
552 return 0; | |
553 } | |
554 | |
555 static int sipr_decode_frame(AVCodecContext *avctx, void *datap, | |
556 int *data_size, AVPacket *avpkt) | |
557 { | |
558 SiprContext *ctx = avctx->priv_data; | |
559 const uint8_t *buf=avpkt->data; | |
560 SiprParameters parm; | |
561 GetBitContext gb; | |
562 float *data = datap; | |
563 int i; | |
564 | |
565 ctx->avctx = avctx; | |
566 if (avpkt->size < (ctx->m.bits_per_frame >> 3)) { | |
567 av_log(avctx, AV_LOG_ERROR, | |
568 "Error processing packet: packet size (%d) too small\n", | |
569 avpkt->size); | |
570 | |
571 *data_size = 0; | |
572 return -1; | |
573 } | |
574 if (*data_size < SUBFR_SIZE * ctx->m.subframe_count * sizeof(float)) { | |
575 av_log(avctx, AV_LOG_ERROR, | |
576 "Error processing packet: output buffer (%d) too small\n", | |
577 *data_size); | |
578 | |
579 *data_size = 0; | |
580 return -1; | |
581 } | |
582 | |
583 init_get_bits(&gb, buf, ctx->m.bits_per_frame); | |
584 | |
585 for (i = 0; i < ctx->m.frames_per_packet; i++) { | |
586 decode_parameters(&parm, &gb, &ctx->m); | |
587 decode_frame(ctx, &parm, data); | |
588 | |
589 data += SUBFR_SIZE * ctx->m.subframe_count; | |
590 } | |
591 | |
592 *data_size = ctx->m.frames_per_packet * SUBFR_SIZE * | |
593 ctx->m.subframe_count * sizeof(float); | |
594 | |
595 return ctx->m.bits_per_frame >> 3; | |
596 }; | |
597 | |
598 AVCodec sipr_decoder = { | |
599 "sipr", | |
600 CODEC_TYPE_AUDIO, | |
601 CODEC_ID_SIPR, | |
602 sizeof(SiprContext), | |
603 sipr_decoder_init, | |
604 NULL, | |
605 NULL, | |
606 sipr_decode_frame, | |
607 .long_name = NULL_IF_CONFIG_SMALL("RealAudio SIPR / ACELP.NET"), | |
608 }; |