Line data Source code
1 : /*====================================================================================
2 : EVS Codec 3GPP TS26.452 Aug 12, 2021. Version 16.3.0
3 : ====================================================================================*/
4 : #include <stdint.h>
5 : #include "options.h" /* Compilation switches */
6 : #include "cnst.h" /* Common constants */
7 : //#include "prot_fx.h" /* Function prototypes */
8 : #include "rom_com.h" /* Function prototypes */
9 : #include "prot_fx.h" /* Function prototypes */
10 : #include "prot_fx_enc.h" /* Function prototypes */
11 :
12 : /*-------------------------------------------------------------------*
13 : * Local constants
14 : *-------------------------------------------------------------------*/
15 :
16 : #define L_ENR ( NB_SSF + 2 )
17 :
18 :
19 : /*-------------------------------------------------------------------*
20 : * find_ener_decrease_fx()
21 : *
22 : * Find maximum energy ration between short subblocks in case
23 : * energy is trailing off after a spike
24 : *-------------------------------------------------------------------*/
25 :
26 19853 : static Word16 find_ener_decrease_fx( /* o : maximum energy ratio Q10*/
27 : const Word16 ind_deltaMax, /* i : index of the beginning of maximum energy search Q0*/
28 : const Word32 *pt_enr_ssf /* i : Pointer to the energy buffer Qx*/
29 : )
30 : {
31 : Word16 i, j, end, flag;
32 : Word16 wtmp0, wtmp1;
33 : Word32 maxEnr, minEnr;
34 : Word16 dE2, exp0, exp1;
35 : #ifndef ISSUE_1770_replace_shr_ro
36 : #ifdef BASOP_NOGLOB_DECLARE_LOCAL
37 : Flag Overflow = 0;
38 : move32();
39 : #endif
40 : #endif
41 :
42 19853 : dE2 = 0;
43 19853 : move16();
44 :
45 19853 : j = ind_deltaMax + 2;
46 19853 : move16();
47 19853 : end = j + L_ENR;
48 19853 : move16();
49 19853 : maxEnr = L_add( pt_enr_ssf[j], 0 );
50 19853 : j = add( j, 1 );
51 19853 : flag = 0;
52 19853 : move16();
53 198530 : FOR( i = j; i < end; i++ )
54 : {
55 178677 : test();
56 178677 : IF( ( GT_32( pt_enr_ssf[i], maxEnr ) ) && ( flag == 0 ) )
57 : {
58 7838 : maxEnr = L_add( pt_enr_ssf[i], 0 ); /*Qx*/
59 7838 : j = add( j, 1 );
60 : }
61 : ELSE
62 : {
63 170839 : flag = 1;
64 170839 : move16();
65 : }
66 : }
67 :
68 19853 : minEnr = L_add( maxEnr, 0 );
69 190692 : FOR( i = j; i < end; i++ )
70 : {
71 170839 : minEnr = L_min( minEnr, pt_enr_ssf[i] ); /* Qx */
72 : }
73 :
74 :
75 19853 : minEnr = L_add_sat( minEnr, 100000 );
76 19853 : exp0 = norm_l( minEnr );
77 19853 : wtmp0 = extract_h( L_shl( minEnr, exp0 ) );
78 19853 : exp1 = sub( norm_l( maxEnr ), 1 );
79 19853 : wtmp1 = extract_h( L_shl( maxEnr, exp1 ) );
80 19853 : wtmp1 = div_s( wtmp1, wtmp0 );
81 : #ifdef ISSUE_1770_replace_shr_ro
82 19853 : dE2 = shr_r_sat( wtmp1, add( sub( exp1, exp0 ), 15 - 10 ) ); /*Q10*/
83 : #else
84 : dE2 = shr_ro( wtmp1, add( sub( exp1, exp0 ), 15 - 10 ), &Overflow ); /*Q10*/
85 : #endif
86 :
87 19853 : return dE2;
88 : }
89 :
90 : /*-------------------------------------------------------------------*
91 : * find_uv_fx()
92 : *
93 : * Decision about coder type
94 : *-------------------------------------------------------------------*/
95 :
96 3100 : Word16 find_uv_fx( /* o : coding type */
97 : Encoder_State *st_fx, /* i/o: encoder state structure */
98 : const Word16 *T_op_fr, /* i : pointer to adjusted fractional pitch (4 val.) Q6*/
99 : const Word16 *voicing_fr, /* i : refined correlation for each subframes Q15*/
100 : const Word16 *speech, /* i : pointer to speech signal for E computation Q_new*/
101 : const Word32 *ee, /* i : lf/hf Energy ratio for present frame Q6*/
102 : const Word16 corr_shift, /* i : normalized correlation correction in noise Q15*/
103 : const Word16 relE, /* i : relative frame energy Q8*/
104 : const Word16 Etot, /* i : total energy Q8*/
105 : const Word32 hp_E[], /* i : energy in HF Q_new + Q_SCALE*/
106 : const Word16 Q_new,
107 : Word16 *flag_spitch, /* i/o: flag to indicate very short stable pitch and high correlation Q0*/
108 : const Word16 shift,
109 : const Word16 last_core_orig /* i : original last core Q0*/
110 : )
111 : {
112 : Word16 coder_type, i;
113 : Word32 mean_ee, dE1, fac_32;
114 : const Word16 *pt_speech;
115 : Word32 L_tmp, enr_ssf[2 * NB_SSF + 2 * NB_SSF + 2], E_min_th;
116 : Word16 dE2;
117 : Word16 ind_deltaMax, tmp_offset_flag;
118 : Word32 Ltmp0, *pt_enr_ssf, *pt_enr_ssf1, dE2_th;
119 : Word16 exp0, exp1, Q_in;
120 : Word16 wtmp0, wtmp1;
121 : Word16 fac, mean_voi3, dE3;
122 : Word16 relE_thres;
123 : Word16 mean_voi3_offset;
124 : Word16 voicing_m, dpit1, dpit2, dpit3;
125 : Word16 ee0_th, ee1_th, voi_th, nb_cond, flag_low_relE;
126 3100 : NOISE_EST_HANDLE hNoiseEst = st_fx->hNoiseEst;
127 3100 : SC_VBR_ENC_HANDLE hSC_VBR = st_fx->hSC_VBR;
128 : #ifdef BASOP_NOGLOB_DECLARE_LOCAL
129 3100 : Flag Overflow = 0;
130 3100 : move32();
131 : #endif
132 :
133 : Word16 Last_Resort;
134 : Word16 vadnoise;
135 :
136 3100 : IF( hSC_VBR != NULL )
137 : {
138 3100 : Last_Resort = hSC_VBR->Last_Resort; /* Q0 */
139 3100 : vadnoise = hSC_VBR->vadnoise_fx;
140 3100 : move16();
141 3100 : move16();
142 : }
143 : ELSE
144 : {
145 0 : Last_Resort = 0;
146 0 : vadnoise = 0;
147 0 : move16();
148 0 : move16();
149 : }
150 :
151 3100 : Q_in = sub( Q_new, 1 );
152 :
153 : /*-----------------------------------------------------------------*
154 : * Detect sudden energy increases to catch voice and music
155 : * temporal events (dE1)
156 : *
157 : * - Find maximum energy per short subblocks.
158 : * Two subblock sets are used shifted by half the subblock length
159 : * - Find maximum energy ratio between adjacent subblocks
160 : *-----------------------------------------------------------------*/
161 :
162 : /* Find maximum energy per short subblocks */
163 3100 : pt_speech = speech - SSF; /* Q_new */
164 3100 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
165 58900 : FOR( i = 0; i < 2 * ( NB_SSF + 1 ); i++ )
166 : {
167 55800 : emaximum_fx( Q_in, pt_speech, SSF, pt_enr_ssf );
168 55800 : pt_speech += ( SSF / 2 );
169 55800 : pt_enr_ssf++;
170 : }
171 :
172 3100 : dE1 = 0;
173 3100 : move16();
174 3100 : ind_deltaMax = 0;
175 3100 : move16();
176 3100 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
177 3100 : pt_enr_ssf1 = pt_enr_ssf + 2;
178 :
179 : /* Test on energy increase between adjacent sub-subframes */
180 3100 : exp1 = 0;
181 3100 : move16();
182 52700 : FOR( i = 0; i < 2 * NB_SSF; i++ )
183 : {
184 : /*fac = *pt_enr_ssf1 / (*pt_enr_ssf + 1);*/
185 49600 : Ltmp0 = L_max( *pt_enr_ssf, 1 );
186 49600 : exp0 = norm_l( Ltmp0 );
187 49600 : wtmp0 = extract_h( L_shl( Ltmp0, exp0 ) );
188 49600 : exp1 = sub( norm_l( *pt_enr_ssf1 ), 1 );
189 49600 : wtmp1 = extract_h( L_shl( *pt_enr_ssf1, exp1 ) );
190 49600 : fac = div_s( wtmp1, wtmp0 );
191 49600 : fac_32 = L_shr_o( L_deposit_l( fac ), add( sub( exp1, exp0 ), 15 - 13 ), &Overflow ); /* fac32 in Q13*/
192 :
193 49600 : if ( GT_32( fac_32, dE1 ) )
194 : {
195 10463 : ind_deltaMax = i;
196 10463 : move16();
197 : }
198 :
199 49600 : dE1 = L_max( dE1, fac_32 ); /* Q13 */
200 :
201 49600 : pt_enr_ssf++;
202 49600 : pt_enr_ssf1++;
203 : }
204 :
205 : /*-----------------------------------------------------------------*
206 : * Average spectral tilt
207 : * Average voicing (normalized correlation)
208 : *-----------------------------------------------------------------*/
209 :
210 : /*mean_ee = 1.0f/3.0f * (st->ee_old + ee[0] + ee[1]); */ /* coefficients take into account the position of the window */
211 3100 : mean_ee = L_add_o( L_add_o( st_fx->ee_old_fx, ee[0], &Overflow ), ee[1], &Overflow ); /* Q6 */
212 3100 : mean_ee = Mult_32_16( mean_ee, 10923 /* 1/3 in Q15 */ ); /*Q6*/
213 :
214 : /* mean_voi3 = 1.0f/3.0f * (voicing[0] + voicing[1] + voicing[2]);*/
215 3100 : Ltmp0 = L_mult( st_fx->voicing_fx[0], 10923 /* 1/3 in Q15 */ ); /* Q31 */
216 3100 : Ltmp0 = L_mac( Ltmp0, st_fx->voicing_fx[1], 10923 /* 1/3 in Q15 */ ); /* Q31 */
217 3100 : mean_voi3 = mac_r_sat( Ltmp0, st_fx->voicing_fx[2], 10923 /* 1/3 in Q15 */ ); /*Q15*/
218 : /*-----------------------------------------------------------------*
219 : * Total frame energy difference (dE3)
220 : *-----------------------------------------------------------------*/
221 :
222 3100 : dE3 = sub( Etot, hNoiseEst->Etot_last_fx ); /*Q8*/
223 :
224 : /*-----------------------------------------------------------------*
225 : * Energy decrease after spike (dE2)
226 : *-----------------------------------------------------------------*/
227 :
228 : /* set different thresholds and conditions for NB and WB input */
229 3100 : dE2_th = 30 << 10;
230 3100 : move32();
231 3100 : nb_cond = 1;
232 3100 : move16(); /* no additional condition for WB input */
233 3100 : IF( EQ_16( st_fx->input_bwidth, NB ) )
234 : {
235 0 : dE2_th = 21 << 10;
236 0 : move32();
237 0 : if ( GE_16( add_o( mean_voi3, corr_shift, &Overflow ), 22282 /* 0.68 in Q15 */ ) ) /*( mean_voi3 + corr_shift ) >= 0.68f*/
238 : {
239 0 : nb_cond = 0;
240 0 : move16();
241 : }
242 : }
243 :
244 : /* calcualte maximum energy decrease */
245 3100 : dE2 = 0;
246 3100 : move16(); /* Test on energy decrease after an energy spike */
247 3100 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
248 :
249 3100 : test();
250 3100 : IF( GT_32( dE1, 30 << 13 ) && nb_cond ) /*>30 Q13*/
251 : {
252 84 : IF( LT_16( sub( shl( NB_SSF, 1 ), ind_deltaMax ), L_ENR ) )
253 : {
254 61 : st_fx->old_ind_deltaMax = ind_deltaMax; /* Q0 */
255 61 : move16();
256 61 : Copy32( pt_enr_ssf, st_fx->old_enr_ssf_fx, 2 * NB_SSF ); /* Qx */
257 : }
258 : ELSE
259 : {
260 23 : st_fx->old_ind_deltaMax = -1;
261 23 : move16();
262 23 : dE2 = find_ener_decrease_fx( ind_deltaMax, pt_enr_ssf ); /*Q10*/
263 :
264 23 : if ( GT_32( dE2, dE2_th ) )
265 : {
266 2 : st_fx->spike_hyst = 0;
267 2 : move16();
268 : }
269 : }
270 : }
271 : ELSE
272 : {
273 3016 : IF( st_fx->old_ind_deltaMax >= 0 )
274 : {
275 59 : Copy32( st_fx->old_enr_ssf_fx, enr_ssf, 2 * NB_SSF ); /* Qx */
276 59 : dE2 = find_ener_decrease_fx( st_fx->old_ind_deltaMax, enr_ssf ); /* Q10 */
277 :
278 59 : if ( GT_32( dE2, dE2_th ) )
279 : {
280 3 : st_fx->spike_hyst = 1;
281 3 : move16();
282 : }
283 : }
284 :
285 3016 : st_fx->old_ind_deltaMax = -1;
286 3016 : move16();
287 : }
288 :
289 : /*-----------------------------------------------------------------*
290 : * Detection of voiced offsets (tmp_offset_flag)
291 : *-----------------------------------------------------------------*/
292 :
293 3100 : tmp_offset_flag = 1;
294 3100 : move16();
295 :
296 3100 : IF( NE_16( st_fx->input_bwidth, NB ) )
297 : {
298 3100 : ee0_th = 154; /*2.4 in Q6 */
299 3100 : move16();
300 3100 : voi_th = 24248; /*0.74f Q15 */
301 3100 : move16();
302 : }
303 : ELSE
304 : {
305 0 : ee0_th = 627; /*9.8f Q6 */
306 0 : move16();
307 0 : voi_th = 24904; /*0.76f Q15*/
308 0 : move16();
309 : }
310 :
311 3100 : test();
312 3100 : test();
313 3100 : test();
314 6069 : if ( ( EQ_16( st_fx->last_coder_type_raw, UNVOICED ) ) || /* previous frame was unvoiced */
315 3989 : ( ( LT_32( ee[0], ee0_th ) ) && ( GT_32( hp_E[0], L_shl( E_MIN_FX, Q_new ) ) ) && /* energy is concentrated in high frequencies provided that some energy is present in HF */
316 485 : ( LT_16( add_o( st_fx->voicing_fx[0], corr_shift, &Overflow ), voi_th ) ) ) ) /* normalized correlation is low */
317 : {
318 312 : tmp_offset_flag = 0;
319 312 : move16();
320 : }
321 :
322 : /*-----------------------------------------------------------------*
323 : * Decision about UC
324 : *-----------------------------------------------------------------*/
325 :
326 : /* SC-VBR - set additional parameters and thresholds for SC-VBR */
327 3100 : mean_voi3_offset = 0;
328 3100 : move16();
329 3100 : flag_low_relE = 0;
330 3100 : move16();
331 3100 : ee1_th = 608; /*9.5 Q6*/
332 3100 : move16();
333 3100 : IF( st_fx->Opt_SC_VBR || ( EQ_16( st_fx->idchan, 1 ) && EQ_16( st_fx->element_mode, IVAS_CPE_TD ) ) ) /* Allow the low energy flag for the secondary channel */
334 : {
335 0 : ee1_th = 544; /*8.5f Q6*/
336 0 : move16();
337 :
338 : /* SC-VBR - determine the threshold on relative energy as a function of lp_noise */
339 0 : IF( NE_16( st_fx->input_bwidth, NB ) )
340 : {
341 : /*relE_thres = 0.700f * st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16) */
342 0 : L_tmp = L_mac( -562036736 /* 33.5f in Q24 */, 22938 /* 0.7 in Q15 */, st_fx->lp_noise_fx ); // Q24
343 0 : if ( Last_Resort == 0 )
344 : {
345 : /*relE_thres = 0.650f * st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16)*/
346 0 : L_tmp = L_mac( -562036736 /* 33.5f in Q24 */, 21299 /* 0.650f in Q15 */, st_fx->lp_noise_fx ); // Q24
347 : }
348 0 : relE_thres = round_fx( L_tmp );
349 : }
350 : ELSE
351 : {
352 :
353 : /*relE_thres = 0.60f * st->lp_noise - 28.2f; (lp_noise in Q8, constant Q8<<16)*/
354 0 : L_tmp = L_mac( -473117491 /* 28.2f in Q24 */, 19661 /* 0.6f in Q15 */, st_fx->lp_noise_fx ); // Q24
355 0 : relE_thres = round_fx( L_tmp );
356 : }
357 0 : relE_thres = s_max( relE_thres, -6400 /* -25.0f in Q8 */ ); /* Q8 */
358 :
359 : /* SC-VBR = set flag on low relative energy */
360 0 : if ( LT_16( relE, relE_thres ) )
361 : {
362 0 : flag_low_relE = 1;
363 0 : move16();
364 : }
365 :
366 : /* SC-VBR - correction of voicing threshold for NB inputs (important only in noisy conditions) */
367 0 : test();
368 0 : if ( EQ_16( st_fx->input_bwidth, NB ) && LT_16( vadnoise, 20 << 8 ) ) /* vadnoise in Q8, constant Q0<<8 */
369 : {
370 0 : mean_voi3_offset = 1638; /*0.05f Q15*/
371 0 : move16();
372 : }
373 : }
374 :
375 : /* make decision whether frame is unvoiced */
376 3100 : E_min_th = L_shl( E_MIN_FX, Q_new );
377 3100 : coder_type = GENERIC;
378 3100 : move16();
379 3100 : IF( EQ_16( st_fx->input_bwidth, NB ) )
380 : {
381 0 : test();
382 0 : test();
383 0 : test();
384 0 : test();
385 0 : test();
386 0 : test();
387 0 : test();
388 0 : test();
389 0 : test();
390 0 : test();
391 0 : test();
392 0 : if ( ( ( LT_16( add_o( mean_voi3, corr_shift, &Overflow ), add( 22282 /* 0.68 in Q15 */, mean_voi3_offset ) ) ) && /* normalized correlation low */
393 0 : ( LT_16( add_o( st_fx->voicing_fx[2], corr_shift, &Overflow ), 25887 /* 0.79 in Q15 */ ) ) && /* normalized correlation low on look-ahead - onset detection */
394 0 : ( LT_32( ee[0], 640 /* 10.0f in Q6 */ ) ) && ( GT_32( hp_E[0], E_min_th ) ) && /* energy concentrated in high frequencies provided that some energy is present in HF... */
395 0 : ( LT_32( ee[1], ee1_th ) ) && ( GT_32( hp_E[1], E_min_th ) ) && /* ... biased towards look-ahead to detect onsets */
396 0 : ( tmp_offset_flag == 0 ) && /* Take care of voiced offsets */
397 : /*( st_fx->music_hysteresis_fx == 0 ) &&*/ /* ... and in segment after AUDIO frames */
398 0 : ( LE_32( dE1, 237568 /* 29.0f in Q13 */ ) ) && /* Avoid on sharp energy spikes */
399 0 : ( LE_32( st_fx->old_dE1_fx, 237568 /* 29.0f in Q13 */ ) ) && /* + one frame hysteresis */
400 0 : ( st_fx->spike_hyst < 0 ) ) || /* Avoid after sharp energy spikes followed by decay (e.g. castanets) */
401 : flag_low_relE ) /* low relative frame energy (only for SC-VBR) */
402 : {
403 0 : coder_type = UNVOICED;
404 0 : move16();
405 : }
406 : }
407 : ELSE
408 : {
409 3100 : test();
410 3100 : test();
411 3100 : test();
412 3100 : test();
413 3100 : test();
414 3100 : test();
415 3100 : test();
416 3100 : test();
417 3100 : test();
418 3100 : test();
419 3100 : test();
420 3100 : test();
421 4035 : if ( ( ( LT_16( add_o( mean_voi3, corr_shift, &Overflow ), add( 22774 /* 0.695f in Q15 */, mean_voi3_offset ) ) ) && /* normalized correlation low */
422 1768 : ( LT_16( add_sat( st_fx->voicing_fx[2], corr_shift ), 25887 /* 0.8 in Q15 */ ) ) && /* normalized correlation low on look-ahead - onset detection */
423 1353 : ( LT_32( ee[0], 397 /* 6.2f in Q6 */ ) ) && ( GT_32( hp_E[0], E_min_th ) ) && /* energy concentrated in high frequencies provided that some energy is present in HF... */
424 467 : ( LT_32( ee[1], 397 /* 6.2f in Q6 */ ) ) && ( GT_32( hp_E[1], E_min_th ) ) && /* ... biased towards look-ahead to detect onsets */
425 151 : ( tmp_offset_flag == 0 ) && /* Take care of voiced offsets */
426 : /*( st_fx->music_hysteresis_fx == 0 ) && */ /* ... and in segment after AUDIO frames */
427 292 : ( LE_32( dE1, 245760 /* 30.0f in Q13 */ ) ) && /* Avoid on sharp energy spikes */
428 141 : ( LE_32( st_fx->old_dE1_fx, 245760 /* 30.0f in Q13 */ ) ) && /* + one frame hysteresis */
429 132 : ( st_fx->spike_hyst < 0 ) ) /* Avoid after sharp energy spikes followed by decay (e.g. castanets) */
430 2969 : || ( flag_low_relE && ( LE_32( st_fx->old_dE1_fx, 245760 /* 30.0f in Q13 */ ) ) ) ) /* low relative frame energy (only for SC-VBR) */
431 : {
432 131 : coder_type = UNVOICED;
433 131 : move16();
434 : }
435 : }
436 :
437 : /*-----------------------------------------------------------------*
438 : * Decision about VC
439 : *-----------------------------------------------------------------*/
440 3100 : if ( st_fx->Opt_SC_VBR )
441 : {
442 0 : hSC_VBR->set_ppp_generic = 0;
443 : }
444 3100 : move16();
445 :
446 3100 : test();
447 3100 : test();
448 3100 : IF( EQ_16( st_fx->localVAD, 1 ) && EQ_16( coder_type, GENERIC ) && NE_16( last_core_orig, AMR_WB_CORE ) )
449 : {
450 2864 : dpit1 = abs_s( sub( T_op_fr[1], T_op_fr[0] ) ); // Q6
451 2864 : dpit2 = abs_s( sub( T_op_fr[2], T_op_fr[1] ) ); // Q6
452 2864 : dpit3 = abs_s( sub( T_op_fr[3], T_op_fr[2] ) ); // Q6
453 :
454 2864 : test();
455 2864 : test();
456 2864 : test();
457 2864 : test();
458 2864 : test();
459 2864 : test();
460 2864 : test();
461 2864 : test();
462 2864 : test();
463 2864 : IF( ( GT_16( voicing_fr[0], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 1st sf. */
464 : ( GT_16( voicing_fr[1], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
465 : ( GT_16( voicing_fr[2], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 3st sf. */
466 : ( GT_16( voicing_fr[3], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 4st sf. */
467 : ( GT_32( mean_ee, 256 /* 4.0f in Q6 */ ) ) && /* energy concentrated in low frequencies */
468 : ( LT_16( dpit1, 3 << 6 ) ) &&
469 : ( LT_16( dpit2, 3 << 6 ) ) &&
470 : ( LT_16( dpit3, 3 << 6 ) ) )
471 : {
472 1383 : coder_type = VOICED;
473 1383 : move16();
474 : }
475 1481 : ELSE IF( st_fx->Opt_SC_VBR && EQ_16( st_fx->input_bwidth, NB ) && LT_16( vadnoise, 20 << 8 ) )
476 : {
477 0 : test();
478 0 : test();
479 0 : test();
480 0 : test();
481 0 : test();
482 0 : test();
483 0 : test();
484 0 : IF( GT_16( voicing_fr[0], 8192 /* 0.25 in Q15 */ ) && /* normalized correlation high in 1st sf. */
485 : ( GT_16( voicing_fr[1], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
486 : ( GT_16( voicing_fr[2], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 3st sf. */
487 : ( GT_16( voicing_fr[3], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 4st sf. */
488 : ( GT_32( mean_ee, 64 ) ) && /* energy concentrated in low frequencies */
489 : ( LT_16( dpit1, 5 << 6 ) ) &&
490 : ( LT_16( dpit2, 5 << 6 ) ) &&
491 : ( LT_16( dpit3, 5 << 6 ) ) )
492 : {
493 0 : hSC_VBR->set_ppp_generic = 1;
494 0 : move16();
495 0 : coder_type = VOICED;
496 0 : move16();
497 : }
498 : }
499 :
500 : /* set VOICED mode for frames with very stable pitch and high correlation
501 : and avoid to switch to AUDIO/MUSIC later */
502 2864 : voicing_m = mac_r( L_mac( L_mac( L_mult( voicing_fr[3], 8192 /* 0.25 in Q15 */ ), voicing_fr[2], 8192 /* 0.25 in Q15 */ ), voicing_fr[1], 8192 /* 0.25 in Q15 */ ), voicing_fr[0], 8192 /* 0.25 in Q15 */ );
503 2864 : test();
504 2864 : test();
505 2864 : test();
506 2864 : test();
507 2864 : test();
508 2864 : IF( *flag_spitch || ( LE_16( dpit1, 3 << 6 ) && LE_16( dpit2, 3 << 6 ) && LE_16( dpit3, 3 << 6 ) &&
509 : GT_16( voicing_m, 31130 /* 0.95f in Q15 */ ) && GT_16( st_fx->voicing_sm_fx, 31785 /* 0.97f in Q15 */ ) ) )
510 : {
511 48 : coder_type = VOICED;
512 48 : move16();
513 48 : *flag_spitch = 1;
514 48 : move16(); /*to avoid switch to AUDIO/MUSIC later*/
515 : }
516 : }
517 :
518 : /*-----------------------------------------------------------------*
519 : * Channel-aware mode - set RF mode and total bitrate
520 : *-----------------------------------------------------------------*/
521 :
522 3100 : st_fx->rf_mode = st_fx->Opt_RF_ON; /* Q0 */
523 3100 : move16();
524 :
525 3100 : IF( EQ_16( coder_type, GENERIC ) )
526 : {
527 1586 : test();
528 1586 : test();
529 1586 : test();
530 1586 : test();
531 1586 : IF( ( LT_16( voicing_fr[0], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
532 : ( LT_16( voicing_fr[1], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
533 : ( LT_16( voicing_fr[2], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 3rd sf. */
534 : ( LT_16( voicing_fr[3], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 4th sf. */
535 : ( GT_16( vadnoise, 25 << 8 ) ) ) /* when speech is clean */
536 :
537 : {
538 0 : st_fx->rf_mode = 0;
539 0 : move16();
540 : /* Current frame cannot be compressed to pack the partial redundancy;*/
541 :
542 0 : if ( st_fx->rf_mode != st_fx->Opt_RF_ON )
543 : {
544 0 : core_coder_mode_switch_fx( st_fx, st_fx->last_total_brate, 0, shift );
545 : }
546 : }
547 : }
548 :
549 : /*-----------------------------------------------------------------*
550 : * Updates
551 : *-----------------------------------------------------------------*/
552 :
553 : /* update spike hysteresis parameters */
554 3100 : test();
555 3100 : if ( st_fx->spike_hyst >= 0 && LT_16( st_fx->spike_hyst, 2 ) )
556 : {
557 7 : st_fx->spike_hyst = add( st_fx->spike_hyst, 1 ); /* Q0 */
558 : }
559 :
560 : /* reset spike hysteresis */
561 3100 : test();
562 3100 : test();
563 3100 : test();
564 3146 : if ( ( GT_16( st_fx->spike_hyst, 1 ) ) &&
565 89 : ( GT_16( dE3, 5 << 8 ) || /* energy increases */
566 61 : ( GT_16( relE, -3328 ) && ( GT_16( add_sat( mean_voi3, corr_shift ), 22774 ) ) ) ) ) /* normalized correlation is high */
567 : {
568 5 : st_fx->spike_hyst = -1;
569 5 : move16();
570 : }
571 :
572 : /* update tilt parameters */
573 3100 : st_fx->ee_old_fx = ee[1];
574 3100 : move32(); /*Q6*/
575 3100 : st_fx->old_dE1_fx = dE1;
576 3100 : move32(); /*Q13*/
577 :
578 : /* save the raw coder_type for various modules later in the codec (the reason is that e.g. UNVOICED is lost at higher rates) */
579 3100 : st_fx->coder_type_raw = coder_type;
580 3100 : move16();
581 :
582 3100 : return coder_type;
583 : }
584 :
585 : /*-------------------------------------------------------------------*
586 : * find_uv()
587 : *
588 : * Decision about coder type
589 : *-------------------------------------------------------------------*/
590 1111194 : Word16 find_uv_ivas_fx( /* o : coding type */
591 : Encoder_State *st_fx, /* i/o: encoder state structure */
592 : const Word16 *T_op_fr, /* i : pointer to adjusted fractional pitch (4 val.) Q6*/
593 : const Word16 *voicing_fr, /* i : refined correlation for each subframes Q15*/
594 : const Word16 *speech, /* i : pointer to speech signal for E computation Q_new*/
595 : const Word32 *ee, /* i : lf/hf Energy ratio for present frame Q6*/
596 : Word32 *dE1X, /* o : sudden energy increase for S/M classifier Q13*/
597 : const Word16 corr_shift, /* i : normalized correlation correction in noise Q15*/
598 : const Word16 relE, /* i : relative frame energy Q8*/
599 : const Word16 Etot, /* i : total energy Q8*/
600 : const Word32 hp_E[], /* i : energy in HF q_hp_E*/
601 : Word16 *flag_spitch, /* i/o: flag to indicate very short stable pitch and high correlation Q0*/
602 : const Word16 last_core_orig, /* i : original last core Q0*/
603 : STEREO_CLASSIF_HANDLE hStereoClassif, /* i/o: stereo classifier structure */
604 : const Word16 Q_new,
605 : const Word16 q_hp_E )
606 : {
607 : Word16 coder_type, i;
608 : Word32 mean_ee, dE1, fac_32;
609 : const Word16 *pt_speech;
610 : Word32 L_tmp, enr_ssf[2 * NB_SSF + 2 * NB_SSF + 2], E_min_th;
611 : Word16 dE2;
612 : Word16 ind_deltaMax, tmp_offset_flag;
613 : Word32 Ltmp0, *pt_enr_ssf, *pt_enr_ssf1, dE2_th;
614 : Word16 exp0, exp1;
615 : Word16 wtmp0, wtmp1;
616 : Word16 fac, mean_voi3, dE3;
617 : Word16 relE_thres;
618 : Word16 mean_voi3_offset;
619 : Word16 voicing_m, dpit1, dpit2, dpit3;
620 : Word16 ee0_th, ee1_th, voi_th, nb_cond, flag_low_relE;
621 1111194 : NOISE_EST_HANDLE hNoiseEst = st_fx->hNoiseEst;
622 1111194 : SC_VBR_ENC_HANDLE hSC_VBR = st_fx->hSC_VBR;
623 : #ifdef BASOP_NOGLOB_DECLARE_LOCAL
624 1111194 : Flag Overflow = 0;
625 1111194 : move32();
626 : #endif
627 : Word16 Last_Resort;
628 : Word16 vadnoise;
629 :
630 1111194 : IF( hSC_VBR != NULL )
631 : {
632 0 : Last_Resort = hSC_VBR->Last_Resort; /* Q0 */
633 0 : move16();
634 0 : vadnoise = hSC_VBR->vadnoise_fx; /* Q8 */
635 0 : move16();
636 : }
637 : ELSE
638 : {
639 1111194 : Last_Resort = 0;
640 1111194 : move16();
641 1111194 : vadnoise = 0;
642 1111194 : move16();
643 : }
644 :
645 : /*-----------------------------------------------------------------*
646 : * Detect sudden energy increases to catch voice and music
647 : * temporal events (dE1)
648 : *
649 : * - Find maximum energy per short subblocks.
650 : * Two subblock sets are used shifted by half the subblock length
651 : * - Find maximum energy ratio between adjacent subblocks
652 : *-----------------------------------------------------------------*/
653 :
654 : /* Find maximum energy per short subblocks */
655 1111194 : pt_speech = speech - SSF;
656 1111194 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
657 21112686 : FOR( i = 0; i < 2 * ( NB_SSF + 1 ); i++ )
658 : {
659 20001492 : emaximum_fx( Q_new, pt_speech, SSF, pt_enr_ssf );
660 20001492 : pt_speech += ( SSF / 2 );
661 20001492 : pt_enr_ssf++;
662 : }
663 :
664 1111194 : dE1 = 0;
665 1111194 : move16();
666 1111194 : ind_deltaMax = 0;
667 1111194 : move16();
668 1111194 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
669 1111194 : pt_enr_ssf1 = pt_enr_ssf + 2;
670 :
671 : /* Test on energy increase between adjacent sub-subframes */
672 1111194 : exp1 = 0;
673 1111194 : move16();
674 18890298 : FOR( i = 0; i < 2 * NB_SSF; i++ )
675 : {
676 : /*fac = *pt_enr_ssf1 / (*pt_enr_ssf + 1);*/
677 17779104 : Ltmp0 = L_max( *pt_enr_ssf, 1 );
678 17779104 : exp0 = norm_l( Ltmp0 );
679 17779104 : wtmp0 = extract_h( L_shl( Ltmp0, exp0 ) );
680 17779104 : exp1 = sub( norm_l( *pt_enr_ssf1 ), 1 );
681 17779104 : wtmp1 = extract_h( L_shl( *pt_enr_ssf1, exp1 ) );
682 17779104 : fac = div_s( wtmp1, wtmp0 );
683 17779104 : fac_32 = L_shr_o( L_deposit_l( fac ), add( sub( exp1, exp0 ), 15 - 13 ), &Overflow ); /* fac32 in Q13*/
684 :
685 17779104 : if ( GT_32( fac_32, dE1 ) )
686 : {
687 3433463 : ind_deltaMax = i;
688 3433463 : move16();
689 : }
690 :
691 17779104 : dE1 = L_max( dE1, fac_32 ); /* Q13 */
692 :
693 17779104 : pt_enr_ssf++;
694 17779104 : pt_enr_ssf1++;
695 : }
696 :
697 1111194 : IF( hStereoClassif != NULL )
698 : {
699 760851 : IF( st_fx->idchan == 0 )
700 : {
701 410255 : hStereoClassif->dE1_ch1_fx = dE1; /* Q13 */
702 410255 : move32();
703 410255 : hStereoClassif->dE1_ch1_e = 31 - Q13;
704 410255 : move16();
705 : }
706 : ELSE
707 : {
708 350596 : hStereoClassif->dE1_ch2_fx = dE1; /* Q13 */
709 350596 : move32();
710 350596 : hStereoClassif->dE1_ch2_e = 31 - Q13;
711 350596 : move16();
712 : }
713 : }
714 :
715 1111194 : if ( dE1X != NULL )
716 : {
717 1111194 : *dE1X = dE1; /* Q13 */
718 1111194 : move32();
719 : }
720 :
721 : /*-----------------------------------------------------------------*
722 : * Average spectral tilt
723 : * Average voicing (normalized correlation)
724 : *-----------------------------------------------------------------*/
725 :
726 : /*mean_ee = 1.0f/3.0f * (st->ee_old + ee[0] + ee[1]); */ /* coefficients take into account the position of the window */
727 1111194 : mean_ee = L_add_o( L_add_o( st_fx->ee_old_fx, ee[0], &Overflow ), ee[1], &Overflow ); /* Q6 */
728 1111194 : mean_ee = Mult_32_16( mean_ee, 10923 ); /*Q6*/
729 :
730 : /* mean_voi3 = 1.0f/3.0f * (voicing[0] + voicing[1] + voicing[2]);*/
731 1111194 : Ltmp0 = L_mult( st_fx->voicing_fx[0], 10923 /* 1/3 in Q15 */ ); /* Q31 */
732 1111194 : Ltmp0 = L_mac( Ltmp0, st_fx->voicing_fx[1], 10923 /* 1/3 in Q15 */ ); /* Q31 */
733 1111194 : mean_voi3 = mac_r_sat( Ltmp0, st_fx->voicing_fx[2], 10923 /* 1/3 in Q15 */ ); /*Q15*/
734 : /*-----------------------------------------------------------------*
735 : * Total frame energy difference (dE3)
736 : *-----------------------------------------------------------------*/
737 :
738 1111194 : dE3 = sub( Etot, extract_h( hNoiseEst->Etot_last_32fx ) ); /*Q8*/
739 :
740 : /*-----------------------------------------------------------------*
741 : * Energy decrease after spike (dE2)
742 : *-----------------------------------------------------------------*/
743 :
744 : /* set different thresholds and conditions for NB and WB input */
745 1111194 : dE2_th = 30 << 10;
746 1111194 : move32();
747 1111194 : nb_cond = 1;
748 1111194 : move16(); /* no additional condition for WB input */
749 1111194 : IF( st_fx->input_bwidth == NB )
750 : {
751 4290 : dE2_th = 21 << 10;
752 4290 : move32();
753 4290 : if ( GE_16( add_o( mean_voi3, corr_shift, &Overflow ), 22282 ) ) /*( mean_voi3 + corr_shift ) >= 0.68f*/
754 : {
755 2880 : nb_cond = 0;
756 2880 : move16();
757 : }
758 : }
759 :
760 : /* calcualte maximum energy decrease */
761 1111194 : dE2 = 0;
762 1111194 : move16(); /* Test on energy decrease after an energy spike */
763 1111194 : pt_enr_ssf = enr_ssf + 2 * NB_SSF;
764 :
765 1111194 : test();
766 1111194 : IF( GT_32( dE1, 30 << 13 ) && nb_cond ) /*>30 Q13*/
767 : {
768 18754 : IF( LT_16( sub( shl( NB_SSF, 1 ), ind_deltaMax ), L_ENR ) )
769 : {
770 7918 : st_fx->old_ind_deltaMax = ind_deltaMax; /* Q0 */
771 7918 : move16();
772 7918 : Copy32( pt_enr_ssf, st_fx->old_enr_ssf_fx, 2 * NB_SSF ); /* Qx */
773 : }
774 : ELSE
775 : {
776 10836 : st_fx->old_ind_deltaMax = -1;
777 10836 : move16();
778 10836 : dE2 = find_ener_decrease_fx( ind_deltaMax, pt_enr_ssf ); /*Q10*/
779 :
780 10836 : if ( GT_32( dE2, dE2_th ) )
781 : {
782 138 : st_fx->spike_hyst = 0;
783 138 : move16();
784 : }
785 : }
786 : }
787 : ELSE
788 : {
789 1092440 : IF( st_fx->old_ind_deltaMax >= 0 )
790 : {
791 8935 : Copy32( st_fx->old_enr_ssf_fx, enr_ssf, 2 * NB_SSF ); /* Qx */
792 8935 : dE2 = find_ener_decrease_fx( st_fx->old_ind_deltaMax, enr_ssf ); /* Q10 */
793 :
794 8935 : if ( GT_32( dE2, dE2_th ) )
795 : {
796 107 : st_fx->spike_hyst = 1;
797 107 : move16();
798 : }
799 : }
800 :
801 1092440 : st_fx->old_ind_deltaMax = -1;
802 1092440 : move16();
803 : }
804 :
805 : /*-----------------------------------------------------------------*
806 : * Detection of voiced offsets (tmp_offset_flag)
807 : *-----------------------------------------------------------------*/
808 :
809 1111194 : tmp_offset_flag = 1;
810 1111194 : move16();
811 :
812 1111194 : IF( st_fx->input_bwidth != NB )
813 : {
814 1106904 : ee0_th = 154; /*2.4 in Q6 */
815 1106904 : move16();
816 1106904 : voi_th = 24248; /*0.74f Q15 */
817 1106904 : move16();
818 : }
819 : ELSE
820 : {
821 4290 : ee0_th = 627; /*9.8f Q6 */
822 4290 : move16();
823 4290 : voi_th = 24904; /*0.76f Q15*/
824 4290 : move16();
825 : }
826 :
827 1111194 : E_min_th = L_shl( E_MIN_IVAS_FX_Q31, sub( q_hp_E, Q31 ) );
828 :
829 1111194 : test();
830 1111194 : test();
831 1111194 : test();
832 1111194 : IF( ( EQ_16( st_fx->last_coder_type_raw, UNVOICED ) ) || /* previous frame was unvoiced */
833 : ( ( LT_32( ee[0], ee0_th ) ) && ( GT_32( hp_E[0], E_min_th ) ) && /* energy is concentrated in high frequencies provided that some energy is present in HF */
834 : ( LT_16( add_o( st_fx->voicing_fx[0], corr_shift, &Overflow ), voi_th ) ) ) ) /* normalized correlation is low */
835 : {
836 127561 : tmp_offset_flag = 0;
837 127561 : move16();
838 : }
839 :
840 : /*-----------------------------------------------------------------*
841 : * Decision about UC
842 : *-----------------------------------------------------------------*/
843 :
844 : /* SC-VBR - set additional parameters and thresholds for SC-VBR */
845 1111194 : mean_voi3_offset = 0;
846 1111194 : move16();
847 1111194 : flag_low_relE = 0;
848 1111194 : move16();
849 1111194 : ee1_th = 608; /*9.5 Q6*/
850 1111194 : move16();
851 1111194 : test();
852 1111194 : test();
853 1111194 : IF( st_fx->Opt_SC_VBR || ( EQ_16( st_fx->idchan, 1 ) && EQ_16( st_fx->element_mode, IVAS_CPE_TD ) ) ) /* Allow the low energy flag for the secondary channel */
854 : {
855 3811 : ee1_th = 544; /*8.5f Q6*/
856 3811 : move16();
857 :
858 : /* SC-VBR - determine the threshold on relative energy as a function of lp_noise */
859 3811 : IF( st_fx->input_bwidth != NB )
860 : {
861 : /*relE_thres = 0.700f * st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16) */
862 3811 : L_tmp = L_mac( -562036736 /* 33.5f in Q24 */, 22938 /* 0.7 in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
863 3811 : IF( Last_Resort == 0 )
864 : {
865 : /*relE_thres = 0.650f * st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16)*/
866 3811 : L_tmp = L_mac( -562036736 /* 33.5f in Q24 */, 21299 /* 0.650f in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
867 : }
868 3811 : relE_thres = round_fx( L_tmp );
869 : }
870 : ELSE
871 : {
872 :
873 : /*relE_thres = 0.60f * st->lp_noise - 28.2f; (lp_noise in Q8, constant Q8<<16)*/
874 0 : L_tmp = L_mac( -473117491 /* 28.2f in Q24 */, 19661 /* 0.6f in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
875 0 : relE_thres = round_fx( L_tmp );
876 : }
877 3811 : relE_thres = s_max( relE_thres, -6400 /* -25.0f in Q8 */ ); /* Q8 */
878 :
879 : /* SC-VBR = set flag on low relative energy */
880 3811 : if ( LT_16( relE, relE_thres ) )
881 : {
882 332 : flag_low_relE = 1;
883 332 : move16();
884 : }
885 :
886 : /* SC-VBR - correction of voicing threshold for NB inputs (important only in noisy conditions) */
887 3811 : test();
888 3811 : if ( st_fx->input_bwidth == NB && LT_16( vadnoise, 20 << 8 ) ) /* vadnoise in Q8, constant Q0<<8 */
889 : {
890 0 : mean_voi3_offset = 1638; /*0.05f Q15*/
891 0 : move16();
892 : }
893 : }
894 :
895 : /* make decision whether frame is unvoiced */
896 1111194 : coder_type = GENERIC;
897 1111194 : move16();
898 1111194 : IF( st_fx->input_bwidth == NB )
899 : {
900 4290 : test();
901 4290 : test();
902 4290 : test();
903 4290 : test();
904 4290 : test();
905 4290 : test();
906 4290 : test();
907 4290 : test();
908 4290 : test();
909 4290 : test();
910 4290 : IF( ( ( LT_16( add_o( mean_voi3, corr_shift, &Overflow ), add( 22282 /* 0.68 in Q15 */, mean_voi3_offset ) ) ) && /* normalized correlation low */
911 : ( LT_16( add_o( st_fx->voicing_fx[2], corr_shift, &Overflow ), 25887 /* 0.79 in Q15 */ ) ) && /* normalized correlation low on look-ahead - onset detection */
912 : ( LT_32( ee[0], 640 /* 10.0f in Q6 */ ) ) && ( GT_32( hp_E[0], E_min_th ) ) && /* energy concentrated in high frequencies provided that some energy is present in HF... */
913 : ( LT_32( ee[1], ee1_th ) ) && ( GT_32( hp_E[1], E_min_th ) ) && /* ... biased towards look-ahead to detect onsets */
914 : ( tmp_offset_flag == 0 ) && /* Take care of voiced offsets */
915 : /*( st_fx->music_hysteresis_fx == 0 ) &&*/ /* ... and in segment after AUDIO frames */
916 : ( LE_32( dE1, 237568 /* 29.0f in Q13 */ ) ) && /* Avoid on sharp energy spikes */
917 : ( LE_32( st_fx->old_dE1_fx, 237568 /* 29.0f in Q13 */ ) ) && /* + one frame hysteresis */
918 : ( st_fx->spike_hyst < 0 ) ) || /* Avoid after sharp energy spikes followed by decay (e.g. castanets) */
919 : flag_low_relE ) /* low relative frame energy (only for SC-VBR) */
920 : {
921 121 : coder_type = UNVOICED;
922 121 : move16();
923 : }
924 : }
925 : ELSE
926 : {
927 1106904 : test();
928 1106904 : test();
929 1106904 : test();
930 1106904 : test();
931 1106904 : test();
932 1106904 : test();
933 1106904 : test();
934 1106904 : test();
935 1106904 : test();
936 1693368 : if ( ( ( LT_16( add_o( mean_voi3, corr_shift, &Overflow ), add( 22774 /* 0.695f in Q15 */, mean_voi3_offset ) ) ) && /* normalized correlation low */
937 : /*( LT_16( add_o( st_fx->voicing_fx[2], corr_shift, &Overflow ), 25887 ) ) && */ /* normalized correlation low on look-ahead - onset detection */
938 960653 : ( LT_32( ee[0], 397 /* 6.2f in Q6 */ ) ) && ( GT_32( hp_E[0], E_min_th ) ) && /* energy concentrated in high frequencies provided that some energy is present in HF... */
939 230620 : ( LT_32( ee[1], 397 /* 6.2f in Q16 */ ) ) && ( GT_32( hp_E[1], E_min_th ) ) && /* ... biased towards look-ahead to detect onsets */
940 82717 : ( tmp_offset_flag == 0 ) && /* Take care of voiced offsets */
941 : /*( st_fx->music_hysteresis_fx == 0 ) && */ /* ... and in segment after AUDIO frames */
942 162186 : ( LE_32( dE1, 245760 /* 30.0f in Q13 */ ) ) && /* Avoid on sharp energy spikes */
943 79469 : ( LE_32( st_fx->old_dE1_fx, 245760 /* 30.0f in Q13 */ ) ) && /* + one frame hysteresis */
944 77100 : ( st_fx->spike_hyst < 0 ) ) /* Avoid after sharp energy spikes followed by decay (e.g. castanets) */
945 1030078 : || ( flag_low_relE && ( LE_32( st_fx->old_dE1_fx, 245760 /* 30.0f in Q13 */ ) ) ) ) /* low relative frame energy (only for SC-VBR) */
946 : {
947 77106 : coder_type = UNVOICED;
948 77106 : move16();
949 : }
950 : }
951 :
952 : /*-----------------------------------------------------------------*
953 : * Decision about VC
954 : *-----------------------------------------------------------------*/
955 1111194 : if ( st_fx->Opt_SC_VBR )
956 : {
957 0 : hSC_VBR->set_ppp_generic = 0;
958 : }
959 1111194 : move16();
960 :
961 1111194 : test();
962 1111194 : test();
963 1111194 : IF( EQ_16( st_fx->localVAD, 1 ) && EQ_16( coder_type, GENERIC ) && NE_16( last_core_orig, AMR_WB_CORE ) )
964 : {
965 873679 : dpit1 = abs_s( sub( T_op_fr[1], T_op_fr[0] ) ); /* Q6 */
966 873679 : dpit2 = abs_s( sub( T_op_fr[2], T_op_fr[1] ) ); /* Q6 */
967 873679 : dpit3 = abs_s( sub( T_op_fr[3], T_op_fr[2] ) ); /* Q6 */
968 :
969 873679 : test();
970 873679 : test();
971 873679 : test();
972 873679 : test();
973 873679 : test();
974 873679 : test();
975 873679 : test();
976 873679 : test();
977 873679 : test();
978 873679 : IF( ( GT_16( voicing_fr[0], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 1st sf. */
979 : ( GT_16( voicing_fr[1], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
980 : ( GT_16( voicing_fr[2], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 3st sf. */
981 : ( GT_16( voicing_fr[3], 19825 /* 0.605 in Q15 */ ) ) && /* normalized correlation high in 4st sf. */
982 : ( GT_32( mean_ee, 256 /* 4.0f in Q6 */ ) ) && /* energy concentrated in low frequencies */
983 : ( LT_16( dpit1, 3 << 6 ) ) &&
984 : ( LT_16( dpit2, 3 << 6 ) ) &&
985 : ( LT_16( dpit3, 3 << 6 ) ) )
986 : {
987 362854 : coder_type = VOICED;
988 362854 : move16();
989 : }
990 510825 : ELSE IF( st_fx->Opt_SC_VBR && st_fx->input_bwidth == NB && LT_16( vadnoise, 20 << 8 ) )
991 : {
992 0 : test();
993 0 : test();
994 0 : test();
995 0 : test();
996 0 : test();
997 0 : test();
998 0 : test();
999 0 : IF( GT_16( voicing_fr[0], 8192 /* 0.25 in Q15 */ ) && /* normalized correlation high in 1st sf. */
1000 : ( GT_16( voicing_fr[1], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
1001 : ( GT_16( voicing_fr[2], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 3st sf. */
1002 : ( GT_16( voicing_fr[3], 8192 /* 0.25 in Q15 */ ) ) && /* normalized correlation high in 4st sf. */
1003 : ( GT_32( mean_ee, 64 /* 1.0f in Q6 */ ) ) && /* energy concentrated in low frequencies */
1004 : ( LT_16( dpit1, 5 << 6 ) ) &&
1005 : ( LT_16( dpit2, 5 << 6 ) ) &&
1006 : ( LT_16( dpit3, 5 << 6 ) ) )
1007 : {
1008 0 : hSC_VBR->set_ppp_generic = 1;
1009 0 : move16();
1010 0 : coder_type = VOICED;
1011 0 : move16();
1012 : }
1013 : }
1014 :
1015 : /* set VOICED mode for frames with very stable pitch and high correlation
1016 : and avoid to switch to AUDIO/MUSIC later */
1017 873679 : voicing_m = mac_r( L_mac( L_mac( L_mult( voicing_fr[3], 8192 /* 0.25 in Q15 */ ), voicing_fr[2], 8192 /* 0.25 in Q15 */ ), voicing_fr[1], 8192 /* 0.25 in Q15 */ ), voicing_fr[0], 8192 /* 0.25 in Q15 */ ); /* Q15 */
1018 873679 : test();
1019 873679 : test();
1020 873679 : test();
1021 873679 : test();
1022 873679 : test();
1023 873679 : IF( *flag_spitch || ( LE_16( dpit1, 3 << 6 ) && LE_16( dpit2, 3 << 6 ) && LE_16( dpit3, 3 << 6 ) &&
1024 : GT_16( voicing_m, 31130 /* 0.95f in Q15 */ ) && GT_16( st_fx->voicing_sm_fx, 31785 /* 0.97f in Q15 */ ) ) )
1025 : {
1026 37531 : coder_type = VOICED;
1027 37531 : move16();
1028 37531 : *flag_spitch = 1;
1029 37531 : move16(); /*to avoid switch to AUDIO/MUSIC later*/
1030 : }
1031 : }
1032 :
1033 : /*-----------------------------------------------------------------*
1034 : * Channel-aware mode - set RF mode and total bitrate
1035 : *-----------------------------------------------------------------*/
1036 :
1037 1111194 : st_fx->rf_mode = st_fx->Opt_RF_ON; /* Q0 */
1038 1111194 : move16();
1039 :
1040 1111194 : IF( EQ_16( coder_type, GENERIC ) )
1041 : {
1042 665994 : test();
1043 665994 : test();
1044 665994 : test();
1045 665994 : test();
1046 665994 : IF( ( LT_16( voicing_fr[0], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
1047 : ( LT_16( voicing_fr[1], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 2st sf. */
1048 : ( LT_16( voicing_fr[2], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 3rd sf. */
1049 : ( LT_16( voicing_fr[3], 6554 /* 0.2f in Q15 */ ) ) && /* normalized correlation high in 4th sf. */
1050 : ( GT_16( vadnoise, 25 << 8 ) ) ) /* when speech is clean */
1051 :
1052 : {
1053 0 : st_fx->rf_mode = 0;
1054 0 : move16();
1055 : /* Current frame cannot be compressed to pack the partial redundancy;*/
1056 :
1057 0 : IF( NE_16( st_fx->rf_mode, st_fx->Opt_RF_ON ) )
1058 : {
1059 0 : core_coder_mode_switch_ivas_fx( st_fx, st_fx->last_total_brate, 0 );
1060 : }
1061 : }
1062 : }
1063 :
1064 : /*-----------------------------------------------------------------*
1065 : * UNCLR classifier
1066 : *-----------------------------------------------------------------*/
1067 :
1068 1111194 : IF( hStereoClassif != NULL )
1069 : {
1070 760851 : test();
1071 760851 : test();
1072 760851 : test();
1073 760851 : test();
1074 760851 : test();
1075 760851 : IF( st_fx->element_mode > EVS_MONO && ( EQ_16( coder_type, GENERIC ) || EQ_16( coder_type, UNVOICED ) || coder_type == INACTIVE || st_fx->localVAD == 0 ) && LT_16( hStereoClassif->unclr_sw_enable_cnt[st_fx->idchan], MAX_UV_CNT ) )
1076 : {
1077 483973 : hStereoClassif->unclr_sw_enable_cnt[st_fx->idchan] = add( hStereoClassif->unclr_sw_enable_cnt[st_fx->idchan], 1 );
1078 483973 : move16();
1079 : }
1080 : ELSE
1081 : {
1082 276878 : hStereoClassif->unclr_sw_enable_cnt[st_fx->idchan] = 0;
1083 276878 : move16();
1084 : }
1085 : }
1086 :
1087 : /*-----------------------------------------------------------------*
1088 : * Updates
1089 : *-----------------------------------------------------------------*/
1090 :
1091 : /* update spike hysteresis parameters */
1092 1111194 : test();
1093 1111194 : if ( st_fx->spike_hyst >= 0 && LT_16( st_fx->spike_hyst, 2 ) )
1094 : {
1095 383 : st_fx->spike_hyst = add( st_fx->spike_hyst, 1 ); /* Q0 */
1096 383 : move16();
1097 : }
1098 :
1099 : /* reset spike hysteresis */
1100 1111194 : test();
1101 1111194 : test();
1102 1111194 : test();
1103 1114409 : if ( ( GT_16( st_fx->spike_hyst, 1 ) ) &&
1104 6294 : ( GT_16( dE3, 5 << 8 ) || /* energy increases */
1105 5195 : ( GT_16( relE, -3328 /* 13 in Q8 */ ) && ( GT_16( add_sat( mean_voi3, corr_shift ), 22774 /* 0.695 in Q15 */ ) ) ) ) ) /* normalized correlation is high */
1106 : {
1107 234 : st_fx->spike_hyst = -1;
1108 234 : move16();
1109 : }
1110 :
1111 : /* update tilt parameters */
1112 1111194 : st_fx->ee_old_fx = ee[1];
1113 1111194 : move32(); /*Q6*/
1114 1111194 : st_fx->old_dE1_fx = dE1;
1115 1111194 : move32(); /*Q13*/
1116 :
1117 : /* save the raw coder_type for various modules later in the codec (the reason is that e.g. UNVOICED is lost at higher rates) */
1118 1111194 : st_fx->coder_type_raw = coder_type; /* Q0 */
1119 1111194 : move16();
1120 :
1121 1111194 : return coder_type;
1122 : }
|
