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 "prot.h"
44 : #include "rom_com.h"
45 : #include "wmc_auto.h"
46 :
47 : /*---------------------------------------------------------------------*
48 : * Function prototypes
49 : *---------------------------------------------------------------------*/
50 :
51 : static void bass_pf_1sf_delay( float *syn, const int16_t *T_sf, const float *gainT_sf, const int16_t l_frame, const int16_t l_subfr, float *bpf_noise_buf, int16_t *gain_factor_param, const int16_t disable_bpf, float *mem_deemph_err, float *lp_ener );
52 :
53 :
54 : /*---------------------------------------------------------------------*
55 : * post_decoder()
56 : *
57 : * Perform post-processing
58 : *---------------------------------------------------------------------*/
59 :
60 2824374 : void post_decoder(
61 : Decoder_State *st, /* i/o: decoder memory state pointer */
62 : float synth_buf[],
63 : const float pit_gain[],
64 : const int16_t pitch[],
65 : float signal_out[],
66 : float *bpf_noise_buf )
67 : {
68 : int16_t L_frame, nb_subfr, i;
69 : float *synth2;
70 : float *synth;
71 : int16_t pfstat_on_previous;
72 : int16_t pitch_gain_adjust[NB_SUBFR16k];
73 : float synth_buf2[NBPSF_PIT_MAX + L_FRAME_MAX + M];
74 : int32_t brate;
75 : float A[M + 1];
76 : float pitch_buf[NB_SUBFR16k];
77 : float tmp;
78 : int16_t L_subfr;
79 :
80 2824374 : L_frame = st->L_frame;
81 2824374 : nb_subfr = st->nb_subfr;
82 2824374 : brate = ( st->core_brate > SID_2k40 ) ? st->total_brate : st->last_active_brate;
83 2824374 : pfstat_on_previous = 0;
84 2824374 : if ( st->hPFstat != NULL )
85 : {
86 707832 : pfstat_on_previous = st->hPFstat->on;
87 707832 : st->hPFstat->on = 0;
88 : }
89 2824374 : set_s( pitch_gain_adjust, st->bpf_gain_param, nb_subfr );
90 2824374 : synth = synth_buf + st->hTcxDec->old_synth_len;
91 2824374 : synth2 = synth_buf2 + NBPSF_PIT_MAX;
92 2824374 : if ( st->hBPF != NULL )
93 : {
94 707832 : mvr2r( st->hBPF->pst_old_syn, synth_buf2, NBPSF_PIT_MAX );
95 : }
96 :
97 2824374 : if ( st->tcxonly )
98 : {
99 : /* High bitrates (48kbps and above), high sampling rates (25.6kHz and above) */
100 :
101 2437284 : mvr2r( synth, synth2, L_frame );
102 :
103 2437284 : if ( pfstat_on_previous )
104 : {
105 : /* Past frame was low-bitrate with formant post-filter */
106 720 : lsp2a_stab( st->lsp_old, A, M );
107 720 : mvr2r( st->hPFstat->mem_pf_in + L_SYN_MEM - M, synth - M, M );
108 720 : L_subfr = st->L_frame / st->nb_subfr;
109 720 : residu( A, M, synth, synth_buf, L_subfr );
110 720 : syn_filt( A, M, synth_buf, synth2, L_subfr, st->hPFstat->mem_stp + L_SYN_MEM - M, 0 );
111 720 : scale_st( synth, synth2, &st->hPFstat->gain_prec, L_subfr, -1 );
112 720 : blend_subfr2( synth2 + L_subfr / 2, synth + L_subfr / 2, synth2 + L_subfr / 2 );
113 : }
114 : }
115 : else
116 : {
117 : /* Low bitrates (32kbps and below), low sampling rates (12.8kHz and 16kHz) */
118 387090 : if ( st->last_bwidth == NB )
119 : {
120 : /* NB Post-filter (pitch+formant post-filter) */
121 354 : mvr2r( synth, synth_buf, L_frame );
122 354 : tmp = synth[-1];
123 354 : preemph( synth_buf, st->preemph_fac, L_frame, &tmp );
124 :
125 354 : tmp = 0.0f;
126 1947 : for ( i = 0; i < nb_subfr; i++ )
127 : {
128 1593 : pitch_buf[i] = pitch[i];
129 : }
130 :
131 354 : if ( pfstat_on_previous == 0 )
132 : {
133 354 : st->hPFstat->reset = 1;
134 : }
135 :
136 354 : if ( st->bwidth == NB )
137 : {
138 0 : st->hPFstat->on = 1;
139 0 : nb_post_filt( L_frame, L_SUBFR, st->hPFstat, &tmp, 0, synth_buf, st->mem_Aq, pitch_buf, GENERIC, st->BER_detect, st->lp_noise > LP_NOISE_THRESH ? 1 : ( ( st->core != ACELP_CORE ) || ( st->coder_type == UNVOICED ) ) );
140 : }
141 : else
142 : {
143 354 : st->hPFstat->on = 0;
144 354 : nb_post_filt( L_frame, L_SUBFR, st->hPFstat, &tmp, 0, synth_buf, st->mem_Aq, pitch_buf, AUDIO, st->BER_detect, st->lp_noise > LP_NOISE_THRESH ? 1 : ( ( st->core != ACELP_CORE ) || ( st->coder_type == UNVOICED ) ) );
145 : }
146 :
147 :
148 354 : mvr2r( synth_buf, synth2, L_frame );
149 354 : tmp = synth2[-1];
150 354 : deemph( synth2, st->preemph_fac, L_frame, &tmp );
151 : }
152 : else
153 : {
154 : /* Formant Post-filter */
155 386736 : if ( pfstat_on_previous == 0 )
156 : {
157 325578 : st->hPFstat->reset = 1;
158 : }
159 :
160 386736 : if ( st->bwidth >= WB )
161 : {
162 386736 : st->hPFstat->on = 1;
163 386736 : formant_post_filt( st->hPFstat, synth, st->mem_Aq, synth2, L_frame, L_SUBFR, st->lp_noise, brate, 0 );
164 : }
165 : else
166 : {
167 0 : st->hPFstat->on = 0;
168 0 : formant_post_filt( st->hPFstat, synth, st->mem_Aq, synth2, L_frame, L_SUBFR, st->lp_noise, brate, 1 );
169 : }
170 : }
171 :
172 : /*Bass Post-filter */
173 387090 : bass_pf_1sf_delay( synth2, pitch, pit_gain, L_frame, L_SUBFR, bpf_noise_buf, pitch_gain_adjust, ( st->lp_noise > LP_NOISE_THRESH && st->narrowBand ) ? 1 : 0, &( st->hBPF->pst_mem_deemp_err ), &( st->hBPF->pst_lp_ener ) );
174 : }
175 :
176 : /* Output */
177 2824374 : mvr2r( synth2, signal_out, L_frame );
178 :
179 : /* Update synth2 memory */
180 2824374 : if ( st->hBPF != NULL )
181 : {
182 707832 : mvr2r( synth_buf2 + L_frame, st->hBPF->pst_old_syn, NBPSF_PIT_MAX );
183 : }
184 :
185 2824374 : return;
186 : }
187 :
188 :
189 : /*---------------------------------------------------------------------*
190 : * bass_pf_1sf_delay()
191 : *
192 : * Perform low-frequency postfiltering
193 : *---------------------------------------------------------------------*/
194 :
195 387090 : static void bass_pf_1sf_delay(
196 : float *syn, /* i : synthesis to postfilter */
197 : const int16_t *T_sf, /* i : Pitch period for all subframes (T_sf[4]) */
198 : const float *gainT_sf, /* i : Pitch gain for all subframes (gainT_sf[4]) */
199 : const int16_t L_frame, /* i : frame length (multiple of l_subfr) */
200 : const int16_t L_subfr_in, /* i : sub-frame length (80/64) */
201 : float *bpf_noise_buf, /* i : harmoninc filtered signal */
202 : int16_t *gain_factor_param, /* i : gain factor param 0-> minimum BPF, 3-> full BPF */
203 : const int16_t disable_bpf, /* i : flag to disable BPF */
204 : float *mem_deemph_err, /* i/o: Error deemphasis memory */
205 : float *lp_ener /* i/o: long_term error signal energy */
206 : )
207 : {
208 : int16_t i, sf, i_subfr, T, lg, L_subfr;
209 : float tmp, corr, ener, gain;
210 : float noise_buf[( 2 * L_SUBFR )], *noise_in;
211 : float error[L_SUBFR];
212 : float ener2;
213 :
214 387090 : noise_in = noise_buf;
215 :
216 387090 : sf = 0;
217 387090 : L_subfr = L_subfr_in;
218 :
219 2160492 : for ( i_subfr = 0; i_subfr < L_frame; i_subfr += L_subfr, sf++ )
220 : {
221 1773402 : if ( i_subfr == 0 )
222 : {
223 387090 : L_subfr = L_subfr_in;
224 : }
225 1386312 : else if ( i_subfr == L_frame )
226 : {
227 0 : L_subfr = 0;
228 : }
229 : else
230 : {
231 1386312 : L_subfr = L_subfr_in;
232 : }
233 :
234 1773402 : T = T_sf[sf];
235 1773402 : gain = gainT_sf[sf];
236 :
237 1773402 : if ( gain > 1.0f )
238 : {
239 1227 : gain = 1.0f;
240 : }
241 1773402 : if ( gain < 0.0f )
242 : {
243 0 : gain = 0.0f;
244 : }
245 :
246 1773402 : lg = L_frame - T - i_subfr;
247 1773402 : if ( lg < 0 )
248 : {
249 1686 : lg = 0;
250 : }
251 1773402 : if ( lg > L_subfr )
252 : {
253 999273 : lg = L_subfr;
254 : }
255 :
256 1773402 : if ( !disable_bpf && gain > 0 )
257 : {
258 9051 : corr = 0.01f;
259 9051 : ener = 0.01f;
260 :
261 423642 : for ( i = 0; i < lg; i++ )
262 : {
263 414591 : corr += syn[i + i_subfr] * ( 0.5f * syn[i + i_subfr - T] + 0.5f * syn[i + i_subfr + T] );
264 414591 : ener += ( 0.5f * syn[i + i_subfr - T] + 0.5f * syn[i + i_subfr + T] ) * ( 0.5f * syn[i + i_subfr - T] + 0.5f * syn[i + i_subfr + T] );
265 : }
266 :
267 173724 : for ( i = lg; i < L_subfr; i++ )
268 : {
269 164673 : corr += syn[i + i_subfr] * syn[i + i_subfr - T];
270 164673 : ener += syn[i + i_subfr - T] * syn[i + i_subfr - T];
271 : }
272 9051 : gain = corr / ener;
273 :
274 9051 : if ( gain > 1.f )
275 : {
276 3396 : gain = 1.0f;
277 : }
278 5655 : else if ( gain < 0.f )
279 : {
280 258 : gain = 0.f;
281 : }
282 :
283 9051 : ener2 = 0.01f;
284 423642 : for ( i = 0; i < lg; i++ )
285 : {
286 414591 : error[i] = gain * ( syn[i + i_subfr] - 0.5f * syn[i + i_subfr - T] - 0.5f * syn[i + i_subfr + T] );
287 414591 : error[i] = error[i] + 0.9f * *mem_deemph_err;
288 414591 : *mem_deemph_err = error[i];
289 414591 : ener2 += error[i] * error[i];
290 : }
291 :
292 173724 : for ( i = lg; i < L_subfr; i++ )
293 : {
294 164673 : error[i] = 0.5f * gain * ( syn[i + i_subfr] - syn[i + i_subfr - T] );
295 164673 : error[i] = error[i] + 0.9f * *mem_deemph_err;
296 164673 : *mem_deemph_err = error[i];
297 164673 : ener2 += error[i] * error[i];
298 : }
299 :
300 9051 : ener2 = (float) ( 10.f * log10( ener2 ) );
301 9051 : *lp_ener = (float) ( 0.99f * *lp_ener + 0.01f * ener2 );
302 9051 : ener2 = (float) pow( 10.f, 0.1f * *lp_ener );
303 9051 : tmp = 0.5f * corr / ( ener + ener2 );
304 :
305 9051 : if ( tmp > 0.5f )
306 : {
307 471 : tmp = 0.5f;
308 : }
309 8580 : else if ( tmp < 0.f )
310 : {
311 258 : tmp = 0.0f;
312 : }
313 :
314 : /*Adjust gain*/
315 : /* full gain = gainLTP*0.5*/
316 : /* adaptive gain = gainLTP*0.5*gain_factor*0.5*/
317 9051 : tmp *= max( 0.5f * gain_factor_param[sf], 0.125f );
318 :
319 : /* calculate noise based on voiced pitch */
320 423642 : for ( i = 0; i < lg; i++ )
321 : {
322 414591 : noise_in[i] = tmp * ( syn[i + i_subfr] - 0.5f * syn[i + i_subfr - T] - 0.5f * syn[i + i_subfr + T] );
323 : }
324 :
325 173724 : for ( i = lg; i < L_subfr; i++ )
326 : {
327 164673 : noise_in[i] = tmp * ( syn[i + i_subfr] - syn[i + i_subfr - T] );
328 : /*It simulates an extrapolation of the buffer syn: syn[i+i_subfr+T]=syn[i+i_subfr]
329 : * -> reduce nrg of noise_in and avoid too much post-filtering*/
330 : /*noise_in[i] = tmp * (syn[i+i_subfr] - 0.5f*syn[i+i_subfr-T] - 0.5f*syn[i+i_subfr]);*/
331 : /*->noise_in[i] = tmp * 0.5f * (syn[i+i_subfr] - syn[i+i_subfr-T]);*/
332 164673 : noise_in[i] *= 0.5f;
333 : }
334 : }
335 : else
336 : {
337 1764351 : set_zero( noise_in, L_subfr );
338 : }
339 :
340 : /* copy bpf noise signal to buffer */
341 1773402 : mvr2r( noise_in, bpf_noise_buf + i_subfr, L_subfr );
342 : }
343 :
344 387090 : return;
345 : }
346 : /*---------------------------------------------------------------------*
347 : * cldfb_synth_set_bandsToZero()
348 : *
349 : *
350 : *---------------------------------------------------------------------*/
351 :
352 0 : void cldfb_synth_set_bandsToZero(
353 : Decoder_State *st,
354 : float **rAnalysis,
355 : float **iAnalysis,
356 : const int16_t nTimeSlots )
357 : {
358 : float nrg_bwddec, nrg_band[CLDFB_NO_CHANNELS_MAX], thr_bwddwc, max_nrg, realQ1, imagQ1;
359 : int16_t flag, offset, WBcnt, i, k, update_perc;
360 : float perc_detect, perc_miss;
361 :
362 0 : realQ1 = 0.0f;
363 0 : imagQ1 = 0.0f;
364 :
365 0 : set_f( nrg_band, 0.0f, CLDFB_NO_CHANNELS_MAX );
366 0 : max_nrg = 0.0f;
367 :
368 0 : offset = 250;
369 0 : WBcnt = 20;
370 0 : perc_miss = 0.83f;
371 0 : perc_detect = 0.93f;
372 :
373 0 : if ( st->VAD == 1 )
374 : {
375 0 : st->active_frame_cnt_bwddec++;
376 0 : st->total_frame_cnt_bwddec++;
377 0 : if ( st->active_frame_cnt_bwddec > 99 )
378 : {
379 0 : st->active_frame_cnt_bwddec = 100;
380 : }
381 0 : if ( st->total_frame_cnt_bwddec > 500 )
382 : {
383 0 : st->total_frame_cnt_bwddec = 500;
384 : }
385 :
386 0 : for ( i = 0; i < ( st->cldfbSyn->no_channels - st->cldfbSyn->bandsToZero ); i++ )
387 : {
388 0 : nrg_bwddec = 0.0f;
389 0 : for ( k = 0; k < nTimeSlots; k++ )
390 : {
391 0 : realQ1 = rAnalysis[k][i];
392 0 : imagQ1 = iAnalysis[k][i];
393 0 : nrg_bwddec += ( realQ1 * realQ1 );
394 0 : nrg_bwddec += ( imagQ1 * imagQ1 );
395 : }
396 0 : nrg_band[i] = ( nrg_bwddec );
397 0 : if ( ( nrg_band[i] > max_nrg ) && ( i > 11 ) )
398 : {
399 0 : max_nrg = nrg_band[i];
400 : }
401 : }
402 0 : for ( ; i < st->cldfbSyn->no_channels; i++ )
403 : {
404 0 : nrg_band[i] = 0;
405 : }
406 :
407 0 : nrg_bwddec = 0;
408 0 : for ( i = 2; i < 9; i++ )
409 : {
410 0 : nrg_bwddec += ( nrg_band[i] / 7.0f );
411 : }
412 :
413 0 : thr_bwddwc = ( nrg_bwddec / 512.0f );
414 :
415 0 : st->avg_nrg_LT = 0.98999f * st->avg_nrg_LT + 0.009979f * thr_bwddwc;
416 0 : update_perc = 1;
417 0 : if ( st->ini_frame >= 25 && thr_bwddwc < st->avg_nrg_LT * 0.005f )
418 : {
419 0 : update_perc = 0;
420 : }
421 :
422 0 : flag = 1;
423 0 : if ( max_nrg >= thr_bwddwc )
424 : {
425 0 : flag = 0;
426 : }
427 :
428 0 : for ( i = 0; i < WBcnt - 1; i++ )
429 : {
430 0 : st->flag_buffer[i] = st->flag_buffer[i + 1];
431 : }
432 0 : st->flag_buffer[WBcnt - 1] = flag;
433 :
434 : /*long term percentage*/
435 0 : if ( update_perc == 1 )
436 : {
437 0 : st->perc_bwddec += ( flag - st->perc_bwddec ) / st->active_frame_cnt_bwddec;
438 : }
439 0 : if ( ( st->total_frame_cnt_bwddec > offset ) && ( st->active_frame_cnt_bwddec > 50 ) )
440 : {
441 0 : if ( ( st->perc_bwddec >= perc_detect || ( st->perc_bwddec >= perc_miss && st->last_flag_filter_NB ) ) && ( sum_s( st->flag_buffer, WBcnt ) != 0 ) ) /* decision hysterysis */
442 : {
443 0 : st->cldfbSyn->bandsToZero = ( st->cldfbSyn->no_channels - 10 );
444 0 : st->last_flag_filter_NB = 1; /* VAD processing must be dependent on hysterysis, as if hysterysis fails, but threshold passes, we dont want next VAD frames to have NB only */
445 : }
446 : else
447 : {
448 0 : st->last_flag_filter_NB = 0;
449 : }
450 : }
451 : else
452 : {
453 0 : st->last_flag_filter_NB = 0;
454 : }
455 0 : if ( sum_s( st->flag_buffer, WBcnt ) == 0 )
456 : {
457 0 : st->perc_bwddec = 0.0f;
458 0 : st->active_frame_cnt_bwddec = 0;
459 0 : st->total_frame_cnt_bwddec = 0;
460 0 : st->last_flag_filter_NB = 0;
461 : }
462 : }
463 : else
464 : {
465 0 : if ( st->last_flag_filter_NB == 1 )
466 : {
467 0 : st->cldfbSyn->bandsToZero = st->last_active_bandsToZero_bwdec;
468 : }
469 0 : st->total_frame_cnt_bwddec++;
470 0 : if ( st->total_frame_cnt_bwddec > 500 )
471 : {
472 0 : st->total_frame_cnt_bwddec = 500;
473 : }
474 : }
475 :
476 0 : st->last_active_bandsToZero_bwdec = st->cldfbSyn->bandsToZero;
477 :
478 0 : return;
479 : }
|
