Line data Source code
1 : /******************************************************************************************************
2 :
3 : (C) 2022-2025 IVAS codec Public Collaboration with portions copyright Dolby International AB, Ericsson AB,
4 : Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
5 : Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
6 : Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
7 : contributors to this repository. All Rights Reserved.
8 :
9 : This software is protected by copyright law and by international treaties.
10 : The IVAS codec Public Collaboration consisting of Dolby International AB, Ericsson AB,
11 : Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
12 : Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
13 : Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
14 : contributors to this repository retain full ownership rights in their respective contributions in
15 : the software. This notice grants no license of any kind, including but not limited to patent
16 : license, nor is any license granted by implication, estoppel or otherwise.
17 :
18 : Contributors are required to enter into the IVAS codec Public Collaboration agreement before making
19 : contributions.
20 :
21 : This software is provided "AS IS", without any express or implied warranties. The software is in the
22 : development stage. It is intended exclusively for experts who have experience with such software and
23 : solely for the purpose of inspection. All implied warranties of non-infringement, merchantability
24 : and fitness for a particular purpose are hereby disclaimed and excluded.
25 :
26 : Any dispute, controversy or claim arising under or in relation to providing this software shall be
27 : submitted to and settled by the final, binding jurisdiction of the courts of Munich, Germany in
28 : accordance with the laws of the Federal Republic of Germany excluding its conflict of law rules and
29 : the United Nations Convention on Contracts on the International Sales of Goods.
30 :
31 : *******************************************************************************************************/
32 :
33 : /*====================================================================================
34 : EVS Codec 3GPP TS26.443 Nov 04, 2021. Version 12.14.0 / 13.10.0 / 14.6.0 / 15.4.0 / 16.3.0
35 : ====================================================================================*/
36 :
37 : #include <stdint.h>
38 : #include "options.h"
39 : #ifdef DEBUGGING
40 : #include "debug.h"
41 : #endif
42 : #include <math.h>
43 : #include "cnst.h"
44 : #include "rom_com.h"
45 : #include "prot.h"
46 : #include "ivas_prot.h"
47 : #include "wmc_auto.h"
48 :
49 : /*-------------------------------------------------------------------*
50 : * Local function prototypes
51 : *-------------------------------------------------------------------*/
52 :
53 : static float edyn( const float *vec, const int16_t lvec );
54 :
55 :
56 : /*-------------------------------------------------------------------*
57 : * encod_audio()
58 : *
59 : * Encode audio (AC) frames
60 : *-------------------------------------------------------------------*/
61 :
62 15245 : void encod_audio(
63 : Encoder_State *st, /* i/o: State structure */
64 : const float speech[], /* i : input speech */
65 : const float Aw[], /* i : weighted A(z) unquantized for subframes */
66 : const float Aq[], /* i : 12k8 Lp coefficient */
67 : const float *res, /* i : residual signal */
68 : float *synth, /* i/o: core synthesis */
69 : float *exc, /* i/o: current non-enhanced excitation */
70 : float *pitch_buf, /* i/o: floating pitch values for each subframe */
71 : float *voice_factors, /* o : voicing factors */
72 : float *bwe_exc, /* o : excitation for SWB TBE */
73 : const int16_t attack_flag, /* i : attack flag (GSC or TC) */
74 : float *lsf_new, /* i : current frame ISF vector */
75 : float *tmp_noise, /* o : long-term noise energy */
76 : const int16_t tdm_Pitch_reuse_flag, /* i : primary channel pitch reuse flag */
77 : const float tdm_Pri_pitch_buf[] /* i : primary channel pitch buffer */
78 : )
79 : {
80 : const float *p_Aq;
81 : int16_t i, i_subfr, nb_subfr, last_pit_bin;
82 : int16_t T0_tmp, T0_frac_tmp, nb_subfr_flag;
83 : int16_t tmp_nb_bits_tot;
84 : GSC_ENC_HANDLE hGSCEnc;
85 : float Es_pred;
86 : float dct_res[L_FRAME16k], dct_epit[L_FRAME16k];
87 : float m_mean;
88 : float exc_wo_nf[L_FRAME16k];
89 : int16_t nb_bits; /*number of bits*/
90 15245 : BSTR_ENC_HANDLE hBstr = st->hBstr;
91 15245 : LPD_state_HANDLE hLPDmem = st->hLPDmem;
92 :
93 : /*---------------------------------------------------------------*
94 : * Initialization
95 : *---------------------------------------------------------------*/
96 :
97 15245 : hGSCEnc = st->hGSCEnc;
98 :
99 15245 : m_mean = 0.0f;
100 :
101 15245 : T0_tmp = 64;
102 15245 : T0_frac_tmp = 0;
103 15245 : mvr2r( hLPDmem->mem_syn, hGSCEnc->mem_syn_tmp, M );
104 15245 : hGSCEnc->mem_w0_tmp = hLPDmem->mem_w0;
105 15245 : Es_pred = 0;
106 :
107 : /*---------------------------------------------------------------*
108 : * Encode GSC IVAS mode
109 : * Encode GSC attack flag (used to reduce possible pre-echo)
110 : * Encode GSC SWB speech flag
111 : *---------------------------------------------------------------*/
112 :
113 15245 : if ( st->element_mode > EVS_MONO && st->idchan == 0 )
114 : {
115 15239 : push_indice( hBstr, IND_GSC_IVAS_SP, st->GSC_IVAS_mode, 2 );
116 : }
117 :
118 15245 : if ( attack_flag > 0 )
119 : {
120 0 : push_indice( hBstr, IND_GSC_ATTACK, 1, 1 );
121 : }
122 : else
123 : {
124 15245 : push_indice( hBstr, IND_GSC_ATTACK, 0, 1 );
125 : }
126 :
127 15245 : if ( st->GSC_IVAS_mode >= 1 || ( st->coder_type != INACTIVE && ( ( st->element_mode == EVS_MONO && st->total_brate >= ACELP_13k20 ) ||
128 6245 : ( st->element_mode > EVS_MONO && st->total_brate > MIN_BRATE_GSC_NOISY_FLAG && st->bwidth >= SWB && !st->flag_ACELP16k ) ) ) )
129 : {
130 5675 : push_indice( hBstr, IND_GSC_SWB_SPEECH, st->GSC_noisy_speech, 1 );
131 : }
132 :
133 : /*---------------------------------------------------------------*+
134 : * Find and encode the number of subframes
135 : *---------------------------------------------------------------*/
136 :
137 15245 : if ( st->core_brate >= ACELP_9k60 && st->core_brate <= ACELP_13k20 )
138 : {
139 21876 : for ( i = 0; i < 5; i++ )
140 : {
141 18230 : if ( fabs( st->hSpMusClas->gsc_lt_diff_etot[MAX_LT - i - 1] ) > 6.0f && hGSCEnc->cor_strong_limit == 1 )
142 : {
143 103 : hGSCEnc->cor_strong_limit = 0;
144 : }
145 : }
146 : }
147 :
148 15245 : if ( st->GSC_IVAS_mode >= 1 || ( st->GSC_noisy_speech == 1 && st->GSC_IVAS_mode == 0 ) )
149 : {
150 5673 : nb_subfr = NB_SUBFR;
151 5673 : if ( st->GSC_IVAS_mode > 0 && st->GSC_IVAS_mode < 3 && st->core_brate < GSC_L_RATE_STG )
152 : {
153 2043 : nb_subfr = 2;
154 : }
155 :
156 5673 : hGSCEnc->cor_strong_limit = 0;
157 5673 : nb_subfr_flag = 1;
158 : }
159 9572 : else if ( st->L_frame == L_FRAME16k && ( st->core_brate <= ACELP_13k20 || st->coder_type == INACTIVE ) )
160 : {
161 2637 : hGSCEnc->cor_strong_limit = 0;
162 2637 : nb_subfr = SWNB_SUBFR;
163 2637 : nb_subfr_flag = 1;
164 : }
165 : else
166 : {
167 6935 : if ( ( hGSCEnc->cor_strong_limit == 0 || st->coder_type == INACTIVE ) && st->core_brate >= ACELP_9k60 )
168 : {
169 1298 : nb_subfr = 2;
170 1298 : nb_subfr_flag = 0;
171 1298 : hGSCEnc->cor_strong_limit = 0;
172 : }
173 : else
174 : {
175 5637 : nb_subfr = SWNB_SUBFR;
176 5637 : nb_subfr_flag = 1;
177 : }
178 :
179 6935 : if ( st->L_frame == L_FRAME16k && ( st->hSpMusClas->mold_corr > .8f && st->core_brate >= MIN_RATE_4SBFR && st->coder_type != INACTIVE ) )
180 : {
181 0 : nb_subfr *= 2;
182 0 : nb_subfr_flag |= 0x2;
183 : }
184 :
185 6935 : if ( st->L_frame == L_FRAME16k && st->core_brate >= MIN_RATE_4SBFR )
186 : {
187 0 : push_indice( hBstr, IND_HF_NOISE, nb_subfr_flag, 2 );
188 : }
189 6935 : else if ( st->core_brate >= ACELP_9k60 )
190 : {
191 : /* nb_subfr_flag can only have the value 0 or 1 */
192 1305 : push_indice( hBstr, IND_HF_NOISE, nb_subfr_flag, 1 );
193 : }
194 : }
195 :
196 15245 : if ( st->L_frame == L_FRAME16k && nb_subfr == NB_SUBFR )
197 : {
198 0 : nb_subfr = NB_SUBFR16k;
199 : }
200 :
201 : /*---------------------------------------------------------------*
202 : * Compute adaptive (pitch) excitation contribution
203 : *---------------------------------------------------------------*/
204 :
205 15245 : if ( !( st->GSC_IVAS_mode > 0 && st->L_frame / nb_subfr == 2 * L_SUBFR && st->GSC_IVAS_mode < 3 ) &&
206 13202 : ( ( st->core_brate >= MIN_RATE_FCB || st->GSC_noisy_speech ) &&
207 1557 : ( ( nb_subfr == NB_SUBFR && st->L_frame == L_FRAME ) || ( nb_subfr == NB_SUBFR16k && st->L_frame == L_FRAME16k ) ) ) )
208 : {
209 689 : if ( st->element_mode > EVS_MONO )
210 : {
211 689 : nb_bits = 5;
212 : }
213 : else
214 : {
215 0 : nb_bits = Es_pred_bits_tbl[BIT_ALLOC_IDX( st->core_brate, GENERIC, -1, -1 )];
216 : }
217 689 : Es_pred_enc( &Es_pred, &i, st->L_frame, L_SUBFR, res, st->voicing, nb_bits, 0 );
218 689 : push_indice( hBstr, IND_ES_PRED, i, nb_bits );
219 : }
220 :
221 15245 : enc_pit_exc( st, speech, Aw, Aq, Es_pred, res, synth, exc, &T0_tmp, &T0_frac_tmp, pitch_buf, nb_subfr, &hGSCEnc->lt_gpitch, tdm_Pitch_reuse_flag, tdm_Pri_pitch_buf );
222 :
223 : /*---------------------------------------------------------------*
224 : * DCT transform
225 : *---------------------------------------------------------------*/
226 :
227 15245 : edct( exc, dct_epit, st->L_frame, st->element_mode );
228 :
229 15245 : edct( res, dct_res, st->L_frame, st->element_mode );
230 :
231 : /*---------------------------------------------------------------*
232 : * Calculate energy dynamics
233 : *---------------------------------------------------------------*/
234 :
235 137205 : for ( i = 7; i < 15; i++ )
236 : {
237 121960 : m_mean += edyn( dct_res + i * 16, 16 );
238 : }
239 15245 : m_mean *= 0.125f;
240 :
241 15245 : if ( m_mean > hGSCEnc->mid_dyn )
242 : {
243 6895 : hGSCEnc->mid_dyn = 0.2f * hGSCEnc->mid_dyn + 0.8f * m_mean;
244 : }
245 : else
246 : {
247 8350 : hGSCEnc->mid_dyn = 0.6f * hGSCEnc->mid_dyn + 0.4f * m_mean;
248 : }
249 :
250 15245 : if ( st->coder_type != INACTIVE )
251 : {
252 11096 : hGSCEnc->noise_lev = ( NOISE_LEVEL_SP3 + 1 ) - usquant( hGSCEnc->mid_dyn, &m_mean, MIN_DYNAMIC, DYNAMIC_RANGE / GSC_NF_STEPS, GSC_NF_STEPS );
253 11096 : if ( hGSCEnc->noise_lev > NOISE_LEVEL_SP3 )
254 : {
255 5424 : hGSCEnc->noise_lev = NOISE_LEVEL_SP3;
256 : }
257 : }
258 :
259 15245 : hGSCEnc->past_dyn_dec = hGSCEnc->noise_lev;
260 :
261 15245 : if ( st->GSC_IVAS_mode >= 1 )
262 : {
263 4984 : hGSCEnc->noise_lev = NOISE_LEVEL_SP2;
264 4984 : if ( st->GSC_IVAS_mode == 3 ) /* Music like */
265 : {
266 2900 : hGSCEnc->noise_lev = NOISE_LEVEL_SP0;
267 : }
268 2084 : else if ( st->GSC_noisy_speech == 0 )
269 : {
270 1972 : hGSCEnc->noise_lev = NOISE_LEVEL_SP3;
271 : }
272 : }
273 10261 : else if ( st->core_brate <= ACELP_8k00 )
274 : {
275 5083 : if ( hGSCEnc->noise_lev <= NOISE_LEVEL_SP2 )
276 : {
277 2966 : hGSCEnc->noise_lev = NOISE_LEVEL_SP2;
278 : }
279 :
280 5083 : push_indice( hBstr, IND_NOISE_LEVEL, hGSCEnc->noise_lev - NOISE_LEVEL_SP2, 2 );
281 : }
282 5178 : else if ( st->GSC_noisy_speech )
283 : {
284 689 : hGSCEnc->noise_lev = NOISE_LEVEL_SP3;
285 : }
286 : else
287 : {
288 4489 : push_indice( hBstr, IND_NOISE_LEVEL, hGSCEnc->noise_lev - NOISE_LEVEL_SP0, 3 );
289 : }
290 :
291 : /*---------------------------------------------------------------*
292 : * Find and encode the last band where the adaptive (pitch) contribution is significant
293 : *---------------------------------------------------------------*/
294 :
295 15245 : last_pit_bin = Pit_exc_contribution_len( st, dct_res, dct_epit, pitch_buf, &hGSCEnc->pit_exc_hangover );
296 :
297 15245 : if ( last_pit_bin == 0 )
298 : {
299 2242 : hLPDmem->tilt_code = 0.0f;
300 : }
301 : else
302 : {
303 13003 : last_pit_bin++;
304 : }
305 :
306 : /*--------------------------------------------------------------------------------------*
307 : * GSC encoder
308 : *--------------------------------------------------------------------------------------*/
309 :
310 : /* Find the current total number of bits used */
311 15245 : tmp_nb_bits_tot = hBstr->nb_bits_tot;
312 :
313 15245 : if ( st->extl_brate > 0 )
314 : {
315 : /* subtract 1 bit for TBE/BWE BWE flag (bit counted in extl_brate) */
316 15129 : tmp_nb_bits_tot--;
317 : }
318 :
319 15245 : if ( st->coder_type == INACTIVE && st->core_brate <= ACELP_9k60 && st->idchan == 0 )
320 : {
321 : /* add 5 bits for noisiness */
322 350 : tmp_nb_bits_tot += 5;
323 : }
324 :
325 15245 : gsc_enc( st, dct_res, dct_epit, last_pit_bin, tmp_nb_bits_tot, nb_subfr, lsf_new, exc_wo_nf, tmp_noise );
326 :
327 : /*--------------------------------------------------------------------------------------*
328 : * iDCT transform
329 : *--------------------------------------------------------------------------------------*/
330 :
331 15245 : edct( dct_epit, exc, st->L_frame, st->element_mode );
332 :
333 15245 : edct( exc_wo_nf, exc_wo_nf, st->L_frame, st->element_mode );
334 :
335 : /*--------------------------------------------------------------------------------------*
336 : * Remove potential pre-echo in case an onset has been detected
337 : *--------------------------------------------------------------------------------------*/
338 :
339 15245 : pre_echo_att( &hGSCEnc->Last_frame_ener, exc, attack_flag, st->last_coder_type, st->L_frame );
340 :
341 : /*--------------------------------------------------------------------------------------*
342 : * Update BWE excitation
343 : *--------------------------------------------------------------------------------------*/
344 :
345 15245 : if ( st->hBWE_TD != NULL )
346 : {
347 15245 : if ( st->L_frame == L_FRAME16k )
348 : {
349 2637 : set_f( voice_factors, 0.0f, NB_SUBFR16k );
350 2637 : interp_code_4over2( exc, bwe_exc, L_FRAME16k );
351 : }
352 : else
353 : {
354 12608 : set_f( voice_factors, 0.0f, NB_SUBFR );
355 12608 : interp_code_5over2( exc, bwe_exc, L_FRAME );
356 : }
357 : }
358 :
359 : /*--------------------------------------------------------------------------------------*
360 : * Synthesis
361 : *--------------------------------------------------------------------------------------*/
362 :
363 15245 : p_Aq = Aq;
364 78862 : for ( i_subfr = 0; i_subfr < st->L_frame; i_subfr += L_SUBFR )
365 : {
366 63617 : syn_filt( p_Aq, M, &exc_wo_nf[i_subfr], &synth[i_subfr], L_SUBFR, hLPDmem->mem_syn, 1 );
367 63617 : p_Aq += ( M + 1 );
368 : }
369 :
370 : /*--------------------------------------------------------------------------------------*
371 : * Updates
372 : *--------------------------------------------------------------------------------------*/
373 :
374 15245 : hLPDmem->mem_w0 = hGSCEnc->mem_w0_tmp;
375 15245 : mvr2r( exc_wo_nf, exc, st->L_frame );
376 :
377 15245 : return;
378 : }
379 :
380 : /*-------------------------------------------------------------------*
381 : * gsc_enc()
382 : *
383 : * Generic audio signal encoder
384 : *-------------------------------------------------------------------*/
385 :
386 22806 : void gsc_enc(
387 : Encoder_State *st, /* i/o: State structure */
388 : float res_dct_in[], /* i : dct of residual signal */
389 : float exc_dct_in[], /* i/o: dct of pitch-only excitation / total excitation */
390 : const int16_t Diff_len, /* i : Lenght of the difference signal (before pure spectral)*/
391 : const int16_t bits_used, /* i : Number of bit used before frequency Q */
392 : const int16_t nb_subfr, /* i : Number of subframe considered */
393 : const float *lsf_new, /* i : ISFs at the end of the frame */
394 : float *exc_wo_nf, /* o : excitation (in f domain) without noisefill */
395 : float *tmp_noise /* o : long-term noise energy */
396 : )
397 : {
398 : int16_t i;
399 : GSC_ENC_HANDLE hGSCEnc;
400 : BSTR_ENC_HANDLE hBstr;
401 : float exc_diffQ[L_FRAME16k];
402 : float exc_diff[L_FRAME16k];
403 : int16_t bit;
404 : int16_t nb_subbands;
405 : int16_t pvq_len;
406 : int16_t bits_per_bands[MBANDS_GN_BITALLOC16k]; /*Q3*/
407 : int16_t tmp_band;
408 : float concat_in[L_FRAME16k];
409 : float concat_out[L_FRAME16k];
410 : int16_t max_ener_band[MBANDS_GN_BITALLOC16k], j;
411 : float Ener_per_bd_iQ[MBANDS_GN16k];
412 : int16_t last_bin;
413 : int16_t bitallocation_band[MBANDS_GN_BITALLOC16k];
414 : int16_t bitallocation_exc[2];
415 : int16_t npulses[NB_SFM];
416 : int16_t maxpulse[NB_SFM];
417 : float mean_gain;
418 : int16_t seed_init;
419 : float max_eq, max_eq_val;
420 :
421 : /*--------------------------------------------------------------------------------------*
422 : * Initialization
423 : *--------------------------------------------------------------------------------------*/
424 :
425 22806 : hGSCEnc = st->hGSCEnc;
426 22806 : hBstr = st->hBstr;
427 :
428 22806 : bit = bits_used;
429 22806 : if ( st->coder_type == INACTIVE && ( st->tdm_LRTD_flag == 1 || st->element_mode == IVAS_SCE ) && st->core_brate <= GSC_LRES_GAINQ_LIMIT )
430 : {
431 7044 : bit += GSC_LRES_NB_NITS;
432 : }
433 :
434 22806 : set_f( exc_diffQ, 0.0f, st->L_frame );
435 22806 : set_f( Ener_per_bd_iQ, 0.0f, MBANDS_GN16k );
436 :
437 : /*--------------------------------------------------------------------------------------*
438 : * Calculate the difference between the residual spectrum and the spectrum of adaptive excitation
439 : * (non valuable temporal content present in exc_dct_in is already zeroed)
440 : *--------------------------------------------------------------------------------------*/
441 :
442 22806 : v_sub( res_dct_in, exc_dct_in, exc_diff, st->L_frame );
443 22806 : exc_diff[0] = 0;
444 :
445 : /*--------------------------------------------------------------------------------------*
446 : * Multiply the difference spectrum with the normalized spectral shape of the residual signal
447 : * This improves the stability of the differnece spectrum since the spectral shape of the
448 : * residual signal is less suseptible to rapid changes than the difference spectrum
449 : *--------------------------------------------------------------------------------------*/
450 :
451 22806 : if ( Diff_len == 0 )
452 : {
453 2242 : tmp_band = 0;
454 : }
455 : else
456 : {
457 20564 : tmp_band = hGSCEnc->mem_last_pit_band;
458 : }
459 :
460 22806 : Ener_per_band_comp( exc_diff, Ener_per_bd_iQ, MBANDS_GN, 1, st->L_frame );
461 :
462 : /*--------------------------------------------------------------------------------------*
463 : * Gain quantizaion
464 : *--------------------------------------------------------------------------------------*/
465 :
466 22806 : i = 0;
467 106092 : while ( i < SIZE_BRATE_INTERMED_TBL )
468 : {
469 106092 : if ( st->core_brate <= brate_intermed_tbl[i] )
470 : {
471 22806 : break;
472 : }
473 83286 : i++;
474 : }
475 22806 : if ( st->element_mode > EVS_MONO && st->coder_type == AUDIO &&
476 11096 : st->core_brate <= STEREO_GSC_BIT_RATE_ALLOC && brate_intermed_tbl[i] == ACELP_9k60 ) /* Bit allocation should be mapped to 8 kb/s instead of 9.6 kb/s in this case */
477 : {
478 1383 : i--;
479 : }
480 :
481 22806 : mean_gain = gsc_gainQ( hBstr, st->element_mode, st->idchan, Ener_per_bd_iQ, Ener_per_bd_iQ, brate_intermed_tbl[i], st->coder_type, st->bwidth, st->L_frame, st->tdm_LRTD_flag, st->core_brate );
482 :
483 22806 : *tmp_noise = 10.0f * mean_gain;
484 :
485 : /*--------------------------------------------------------------------------------------*
486 : * PVQ encoder
487 : *--------------------------------------------------------------------------------------*/
488 :
489 22806 : bands_and_bit_alloc( hGSCEnc->cor_strong_limit, hGSCEnc->noise_lev, st->core_brate, Diff_len, bit, &bit, Ener_per_bd_iQ, max_ener_band, bits_per_bands, &nb_subbands, exc_diff, concat_in, &pvq_len, st->coder_type, st->bwidth, st->GSC_noisy_speech, st->L_frame, st->element_mode, st->GSC_IVAS_mode );
490 :
491 22806 : set_s( npulses, 0, NB_SFM );
492 :
493 22806 : if ( bit == 0 )
494 : {
495 0 : set_zero( concat_out, L_FRAME16k );
496 : }
497 : else
498 : {
499 22806 : bit -= pvq_core_enc( hBstr, concat_in, concat_out, bit, nb_subbands, gsc_sfm_start, gsc_sfm_end, gsc_sfm_size, bits_per_bands, NULL, npulses, maxpulse, ACELP_CORE );
500 : }
501 :
502 : /* write unused bits */
503 22806 : while ( bit > 0 )
504 : {
505 0 : i = min( bit, 16 );
506 0 : push_indice( hBstr, IND_UNUSED, 0, i );
507 0 : bit -= i;
508 : }
509 :
510 : /* Reorder Q bands */
511 22806 : last_bin = 0;
512 22806 : set_s( bitallocation_band, 0, MBANDS_GN_BITALLOC16k );
513 22806 : seed_init = 0;
514 22806 : max_eq = 0;
515 22806 : max_eq_val = 1.0f;
516 :
517 22806 : if ( ( ( ( st->core_brate < ACELP_7k20 && st->GSC_noisy_speech == 1 ) || st->core_brate < 6000 ) && st->coder_type <= UNVOICED ) || st->GSC_IVAS_mode >= 1 )
518 : {
519 12483 : j = emaximum( concat_out, nb_subbands * 16, &max_eq );
520 12483 : max_eq = max_eq_val / ( fabsf( concat_out[j] ) + 0.01f );
521 12483 : max_eq = min( max_eq_val, max_eq );
522 : }
523 :
524 186633 : for ( j = 0; j < nb_subbands; j++ )
525 : {
526 163827 : mvr2r( concat_out + j * 16, exc_diffQ + max_ener_band[j] * 16, 16 );
527 :
528 163827 : if ( max_ener_band[j] > last_bin )
529 : {
530 118189 : last_bin = max_ener_band[j];
531 : }
532 :
533 163827 : bitallocation_band[max_ener_band[j]] = 1;
534 :
535 163827 : seed_init += npulses[j];
536 : }
537 22806 : if ( st->last_coder_type != AUDIO /* First audio frame */
538 13544 : && st->last_coder_type != UNVOICED ) /* last_coder_type == INACTIVE is overwritten in update_enc to UNVOICED */
539 : {
540 461485 : for ( j = 0; j < nb_subbands * 16; j++ )
541 : {
542 458032 : if ( concat_out[j] > 0 )
543 : {
544 66132 : seed_init = (int16_t) ( (int32_t) seed_init * 8 );
545 : }
546 458032 : if ( concat_out[j] < 0 )
547 : {
548 66269 : seed_init = (int16_t) ( (int32_t) seed_init + 3 );
549 : }
550 : }
551 :
552 3453 : hGSCEnc->seed_tcx = seed_init;
553 : }
554 :
555 22806 : if ( st->core_brate == ACELP_8k00 && st->bwidth != NB )
556 : {
557 44 : bitallocation_exc[0] = 0;
558 44 : bitallocation_exc[1] = 0;
559 :
560 44 : if ( exc_diffQ[L_FRAME8k - 2] != 0 )
561 : {
562 1 : bitallocation_exc[0] = 1;
563 : }
564 :
565 44 : if ( exc_diffQ[L_FRAME8k - 1] != 0 )
566 : {
567 5 : bitallocation_exc[1] = 1;
568 : }
569 : }
570 :
571 : /*--------------------------------------------------------------------------------------*
572 : * Skip adaptive (pitch) contribution frequency band (no noise added over the adaptive (pitch) contribution)
573 : * Find x pulses between 1.6-3.2kHz to code in the spectrum of the residual signal
574 : * Gain is based on the inter-correlation gain between the pulses found and residual signal
575 : *--------------------------------------------------------------------------------------*/
576 :
577 22806 : if ( st->GSC_IVAS_mode >= 1 && st->GSC_noisy_speech == 1 )
578 : {
579 21616 : for ( i = 64; i < st->L_frame; i++ )
580 : {
581 21504 : exc_diffQ[i] *= max_eq;
582 : }
583 : }
584 22694 : else if ( st->core_brate < ACELP_7k20 && st->GSC_noisy_speech == 1 && st->coder_type <= UNVOICED )
585 : {
586 1927243 : for ( i = 0; i < L_FRAME; i++ )
587 : {
588 1919744 : exc_diffQ[i] *= max_eq;
589 : }
590 : }
591 : else
592 : {
593 15195 : freq_dnw_scaling( hGSCEnc->cor_strong_limit, st->coder_type, hGSCEnc->noise_lev, st->core_brate, exc_diffQ, st->L_frame );
594 : }
595 :
596 : /*--------------------------------------------------------------------------------------*
597 : * Estimate noise level
598 : *--------------------------------------------------------------------------------------*/
599 :
600 22806 : highband_exc_dct_in( st->core_brate, mfreq_bindiv_loc, last_bin, Diff_len, hGSCEnc->noise_lev, tmp_band, exc_diffQ, &hGSCEnc->seed_tcx, Ener_per_bd_iQ, nb_subfr, exc_dct_in, st->last_coder_type, bitallocation_band, lsf_new, hGSCEnc->last_exc_dct_in, &hGSCEnc->last_ener, hGSCEnc->last_bitallocation_band, bitallocation_exc, 0, st->coder_type, st->bwidth, exc_wo_nf, st->GSC_noisy_speech, NULL, st->L_frame, st->element_mode, st->GSC_IVAS_mode );
601 :
602 22806 : exc_dct_in[0] = 0;
603 :
604 22806 : return;
605 : }
606 :
607 :
608 : /*---------------------------------------------------------------------*
609 : * edyn()
610 : *
611 : * Calculate energy dynamics in a vector (ratio of energy maximum to energy mean)
612 : *---------------------------------------------------------------------*/
613 :
614 : /*! r: ratio of max to mean */
615 121960 : static float edyn(
616 : const float *vec, /* i : input vector */
617 : const int16_t lvec /* i : length of input vector */
618 : )
619 : {
620 : int16_t j;
621 : float temp, ener_max, ener_mean, dyn;
622 :
623 121960 : ener_mean = 1.0f;
624 121960 : ener_max = 1.0f;
625 :
626 2073320 : for ( j = 0; j < lvec; j++ )
627 : {
628 1951360 : temp = vec[j] * vec[j];
629 :
630 1951360 : if ( temp > ener_max )
631 : {
632 421746 : ener_max = temp;
633 : }
634 1951360 : ener_mean += temp;
635 : }
636 121960 : ener_mean /= lvec;
637 121960 : dyn = 10.0f * ( ener_max / ener_mean );
638 :
639 121960 : return dyn;
640 : }
641 :
642 :
643 : /*-------------------------------------------------------------------*
644 : * GSC_enc_init()
645 : *
646 : * Initialize GSC encoder state structure
647 : *-------------------------------------------------------------------*/
648 :
649 3897 : void GSC_enc_init(
650 : GSC_ENC_HANDLE hGSCEnc /* i/o: GSC data handle */
651 : )
652 : {
653 : /* AC mode */
654 3897 : hGSCEnc->seed_tcx = 15687;
655 3897 : hGSCEnc->cor_strong_limit = 1;
656 3897 : set_f( hGSCEnc->last_exc_dct_in, 0, L_FRAME16k );
657 3897 : hGSCEnc->last_ener = 0.0f;
658 3897 : set_s( hGSCEnc->last_bitallocation_band, 0, 6 );
659 3897 : hGSCEnc->mem_last_pit_band = BAND1k2 + 1;
660 :
661 3897 : hGSCEnc->lt_gpitch = 0.0f;
662 3897 : hGSCEnc->pit_exc_hangover = 0;
663 :
664 : /* GSC - pitch excitation parameters */
665 3897 : hGSCEnc->mem_w0_tmp = 0.0f;
666 3897 : set_f( hGSCEnc->mem_syn_tmp, 0.0f, M );
667 :
668 3897 : hGSCEnc->Last_frame_ener = (float) MAX_32;
669 3897 : hGSCEnc->mid_dyn = 40.0f;
670 3897 : hGSCEnc->noise_lev = NOISE_LEVEL_SP0;
671 3897 : hGSCEnc->past_dyn_dec = 0;
672 :
673 3897 : return;
674 : }
|
