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 "prot.h"
45 : #include "rom_com.h"
46 : #include "wmc_auto.h"
47 :
48 : /*-------------------------------------------------------------------*
49 : * FEC_encode()
50 : *
51 : * Encoder supplementary information for FEC
52 : *-------------------------------------------------------------------*/
53 :
54 157798 : void FEC_encode(
55 : BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
56 : const ACELP_config acelp_cfg, /* i/o: configuration of the ACELP */
57 : const float *synth, /* i : pointer to synthesized speech for E computation */
58 : const int16_t coder_type, /* i : type of coder */
59 : int16_t clas, /* i : signal clas for current frame */
60 : const float *fpit, /* i : close loop fractional pitch buffer */
61 : const float *res, /* i : LP residual signal frame */
62 : int16_t *last_pulse_pos, /* i/o: Position of the last pulse */
63 : const int16_t L_frame, /* i : Frame length */
64 : const int32_t total_brate /* i : total codec bitrate */
65 : )
66 : {
67 : int16_t tmpS, index;
68 157798 : int16_t maxi, sign = 0, tmp_FER_pitch;
69 : float enr_q;
70 :
71 157798 : tmpS = 0;
72 157798 : enr_q = 1.0f;
73 :
74 157798 : if ( coder_type > UNVOICED && coder_type < AUDIO && acelp_cfg.FEC_mode > 0 )
75 : {
76 : /*-----------------------------------------------------------------*
77 : * encode signal class (not needed for VC since it is clearly voiced) (2 bits)
78 : *-----------------------------------------------------------------*/
79 :
80 111939 : if ( coder_type != VOICED )
81 : {
82 : /* encode signal clas with 2 bits */
83 106972 : if ( clas == UNVOICED_CLAS )
84 : {
85 45693 : index = 0;
86 : }
87 61279 : else if ( clas == VOICED_TRANSITION || clas == UNVOICED_TRANSITION )
88 : {
89 7291 : index = 1;
90 : }
91 53988 : else if ( clas == VOICED_CLAS )
92 : {
93 46492 : index = 2;
94 : }
95 : else
96 : {
97 7496 : index = 3;
98 : }
99 :
100 106972 : push_indice( hBstr, IND_FEC_CLAS, index, FEC_BITS_CLS );
101 : }
102 :
103 : /*-----------------------------------------------------------------*
104 : * Encode frame energy (5 bits)
105 : *-----------------------------------------------------------------*/
106 :
107 111939 : if ( acelp_cfg.FEC_mode > 1 ) /* GENERIC and VOICED frames */
108 : {
109 : /* frame energy (maximum energy per pitch period for voiced frames or mean energy per sample over 2nd halframe for unvoiced frames) */
110 63532 : fer_energy( L_frame, clas, synth, fpit[( L_frame >> 6 ) - 1], &enr_q, L_frame );
111 :
112 : /* linearly quantize the energy in the range 0 : FEC_ENR_STEP : 96 dB */
113 63532 : tmpS = (int16_t) ( 10.0 * log10( enr_q + 0.001f ) / FEC_ENR_STEP );
114 :
115 63532 : if ( tmpS > FEC_ENR_QLIMIT )
116 : {
117 0 : tmpS = FEC_ENR_QLIMIT;
118 : }
119 :
120 63532 : if ( tmpS < 0 )
121 : {
122 17 : tmpS = 0;
123 : }
124 :
125 63532 : push_indice( hBstr, IND_FEC_ENR, tmpS, FEC_BITS_ENR );
126 : }
127 :
128 : /*-----------------------------------------------------------------*
129 : * Encode last glottal pulse position (8 bits)
130 : *-----------------------------------------------------------------*/
131 :
132 111939 : if ( acelp_cfg.FEC_mode > 2 ) /* GENERIC frames */
133 : {
134 : /* retrieve the last glottal pulse position of the previous frame */
135 : /* use the current pitch information to scale or not the quantization */
136 10699 : tmp_FER_pitch = (int16_t) ( fpit[0] ); /* take the 1st subframe pitch, since it is easier to retrieve it on decoder side */
137 :
138 10699 : sign = 0;
139 10699 : maxi = *last_pulse_pos;
140 10699 : if ( maxi < 0 )
141 : {
142 1788 : sign = 1;
143 1788 : maxi = -maxi;
144 : }
145 :
146 10699 : if ( tmp_FER_pitch >= 128 )
147 : {
148 3390 : maxi /= 2;
149 : }
150 :
151 10699 : if ( maxi > 127 )
152 : {
153 : /* better not use the glottal pulse position at all instead of using a wrong pulse */
154 : /* can happen only with pitch > 254 and max pit = 289 and should happen very rarely */
155 21 : maxi = 0;
156 : }
157 :
158 10699 : if ( sign == 1 )
159 : {
160 1788 : maxi += 128; /* use 8 bits (MSB represents the sign of the pulse) */
161 : }
162 :
163 10699 : push_indice( hBstr, IND_FEC_POS, maxi, FEC_BITS_POS );
164 : }
165 :
166 : /* find the glottal pulse position of the current frame (could be sent as extra FEC info in the next frame) */
167 111939 : maxi = 0;
168 111939 : if ( clas >= VOICED_CLAS && total_brate >= ACELP_24k40 )
169 : {
170 14458 : maxi = findpulse( L_frame, res, (int16_t) ( fpit[( L_frame >> 6 ) - 1] ), 0, &sign );
171 14458 : if ( sign == 1 )
172 : {
173 5673 : maxi = -maxi;
174 : }
175 : }
176 :
177 111939 : *last_pulse_pos = maxi;
178 : }
179 : else
180 : {
181 45859 : *last_pulse_pos = 0;
182 : }
183 :
184 157798 : return;
185 : }
186 :
187 :
188 : /*-------------------------------------------------------------------*
189 : * FEC_lsf_estim_enc()
190 : *
191 : * Simulates LSF estimation in case of FEC in the encoder ( only one frame erasure is considered )
192 : * The estimated LSF vector is then used to check LSF stability and may invoke safety-net usage in the next frame
193 : *-------------------------------------------------------------------*/
194 :
195 269763 : void FEC_lsf_estim_enc(
196 : Encoder_State *st, /* i : Encoder static memory */
197 : float *lsf /* o : estimated LSF vector */
198 : )
199 : {
200 : int16_t i;
201 : float alpha, lsf_mean[M];
202 :
203 269763 : if ( st->L_frame == L_FRAME )
204 : {
205 120580 : mvr2r( UVWB_Ave, lsf_mean, M );
206 : }
207 : else
208 : {
209 149183 : mvr2r( GEWB2_Ave, lsf_mean, M );
210 : }
211 :
212 : /*----------------------------------------------------------------------*
213 : * Initialize the alpha factor
214 : *----------------------------------------------------------------------*/
215 :
216 269763 : if ( st->last_coder_type == UNVOICED )
217 : {
218 : /* clearly unvoiced */
219 28702 : alpha = ALPHA_UU;
220 : }
221 241061 : else if ( st->last_coder_type == AUDIO || st->clas == INACTIVE_CLAS )
222 : {
223 11361 : alpha = 0.995f;
224 : }
225 229700 : else if ( st->clas == UNVOICED_CLAS )
226 : {
227 : /* if stable, do not flatten the spectrum in the first erased frame */
228 111149 : alpha = st->stab_fac * ( 1.0f - 2.0f * ALPHA_U ) + 2.0f * ALPHA_U;
229 : }
230 118551 : else if ( st->clas == UNVOICED_TRANSITION )
231 : {
232 2759 : alpha = ALPHA_UT;
233 : }
234 115792 : else if ( st->clas == VOICED_CLAS || st->clas == ONSET )
235 : {
236 : /* clearly voiced - mild convergence to the CNG spectrum for the first three erased frames */
237 101608 : alpha = ALPHA_V;
238 : }
239 14184 : else if ( st->clas == SIN_ONSET )
240 : {
241 0 : alpha = ALPHA_S;
242 : }
243 : else
244 : {
245 : /* long erasures and onsets - rapid convergence to the CNG spectrum */
246 14184 : alpha = ALPHA_VT;
247 : }
248 :
249 : /*----------------------------------------------------------------------*
250 : * Extrapolate LSF vector
251 : *----------------------------------------------------------------------*/
252 :
253 : /* extrapolate the old LSF vector */
254 4585971 : for ( i = 0; i < M; i++ )
255 : {
256 : /* calculate mean LSF vector */
257 4316208 : lsf_mean[i] = BETA_FEC * lsf_mean[i] + ( 1 - BETA_FEC ) * st->lsf_adaptive_mean[i];
258 :
259 : /* move old LSF vector towards the mean LSF vector */
260 4316208 : lsf[i] = alpha * st->lsf_old[i] + ( 1.0f - alpha ) * lsf_mean[i];
261 : }
262 :
263 : /* check LSF stability through LSF ordering */
264 269763 : if ( st->L_frame == L_FRAME )
265 : {
266 120580 : reorder_lsf( lsf, MODE1_LSF_GAP, M, INT_FS_12k8 );
267 : }
268 : else /* L_frame == L_FRAME16k */
269 : {
270 149183 : reorder_lsf( lsf, MODE1_LSF_GAP, M, INT_FS_16k );
271 : }
272 :
273 269763 : return;
274 : }
|
