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 <assert.h>
38 : #include <stdint.h>
39 : #include "options.h"
40 : #ifdef DEBUGGING
41 : #include "debug.h"
42 : #endif
43 : #include <math.h>
44 : #include "cnst.h"
45 : #include "prot.h"
46 : #include "wmc_auto.h"
47 :
48 : /*-------------------------------------------------------------------*
49 : * encod_unvoiced()
50 : *
51 : * Encode unvoiced (UC) frames
52 : *-------------------------------------------------------------------*/
53 :
54 2037 : void encod_unvoiced(
55 : Encoder_State *st, /* i/o: state structure */
56 : const float *speech, /* i : input speech */
57 : const float Aw[], /* i : weighted A(z) unquantized for subframes */
58 : const float *Aq, /* i : LP coefficients */
59 : const float Es_pred, /* i : predicted scaled innov. energy */
60 : const int16_t uc_two_stage_flag, /* i : flag indicating two-stage UC */
61 : const float *res, /* i : residual signal */
62 : float *syn, /* o : core synthesis */
63 : float *tmp_noise, /* o : long-term noise energy */
64 : float *exc, /* i/o: current non-enhanced excitation */
65 : float *pitch_buf, /* o : floating pitch values for each subframe */
66 : float *voice_factors, /* o : voicing factors */
67 : float *bwe_exc /* i/o: excitation for SWB TBE */
68 : )
69 : {
70 : float xn[L_SUBFR]; /* Target vector for pitch search */
71 : float h1[L_SUBFR]; /* Impulse response vector */
72 : float code[L_SUBFR]; /* Fixed codebook excitation */
73 : float y2[L_SUBFR]; /* Filtered algebraic excitation */
74 : float *pt_pitch; /* pointer to floating pitch buffer */
75 : float gain_pit; /* Pitch gain */
76 : float voice_fac; /* Voicing factor */
77 : float gain_code; /* gain of code */
78 : float gain_inov; /* inovative gain */
79 : const float *p_Aw, *p_Aq; /* pointer to LP filter coeff. vector */
80 : int16_t i_subfr;
81 : float norm_gain_code;
82 : float cn[L_SUBFR]; /* Target vector in residual domain */
83 : float y1[L_SUBFR]; /* Filtered adaptive excitation */
84 : float code2[L_SUBFR]; /* Gaussian excitation */
85 : float y22[L_SUBFR]; /* Filtered Gaussian excitation */
86 : int16_t i, unbits_PI;
87 : ACELP_CbkCorr g_corr;
88 : float gain_code2, g_corr2[6], exc2[L_SUBFR];
89 : int16_t index;
90 :
91 2037 : LPD_state_HANDLE hLPDmem = st->hLPDmem;
92 :
93 : /*------------------------------------------------------------------*
94 : * Initializations
95 : *------------------------------------------------------------------*/
96 :
97 2037 : gain_pit = 0;
98 :
99 2037 : if ( st->Opt_SC_VBR && st->vad_flag == 0 )
100 : {
101 0 : if ( st->hSC_VBR->last_ppp_mode == 1 || st->hSC_VBR->last_nelp_mode == 1 )
102 : {
103 : /* SC_VBR - reset the encoder, to avoid memory not updated issue for the
104 : case when UNVOICED mode is used to code inactive speech */
105 0 : CNG_reset_enc( st, pitch_buf, voice_factors, 1 );
106 : }
107 : }
108 :
109 2037 : p_Aw = Aw;
110 2037 : p_Aq = Aq;
111 2037 : pt_pitch = pitch_buf;
112 :
113 : /*----------------------------------------------------------------*
114 : * subframe loop
115 : *----------------------------------------------------------------*/
116 :
117 10185 : for ( i_subfr = 0; i_subfr < L_FRAME; i_subfr += L_SUBFR )
118 : {
119 : /*----------------------------------------------------------------*
120 : * Bandwidth expansion of A(z) filter coefficients
121 : * Find the excitation search target "xn" and innovation target in residual domain "cn"
122 : * Compute impulse response, h1[], of weighted synthesis filter
123 : *----------------------------------------------------------------*/
124 :
125 8148 : mvr2r( &res[i_subfr], &exc[i_subfr], L_SUBFR );
126 :
127 8148 : find_targets( speech, hLPDmem->mem_syn, i_subfr, &hLPDmem->mem_w0, p_Aq, res, L_SUBFR, p_Aw, st->preemph_fac, xn, cn, h1 );
128 :
129 8148 : if ( !uc_two_stage_flag )
130 : {
131 : /*----------------------------------------------------------------*
132 : * Unvoiced subframe processing
133 : *----------------------------------------------------------------*/
134 :
135 0 : *pt_pitch = gaus_encode( st, i_subfr, h1, xn, exc, &hLPDmem->mem_w0, st->clip_var, &hLPDmem->tilt_code, code, &gain_code, y2, &gain_inov, &voice_fac, &gain_pit, &norm_gain_code );
136 : }
137 : else
138 : {
139 : /*----------------------------------------------------------------*
140 : * Unvoiced subframe processing in two stages
141 : *----------------------------------------------------------------*/
142 :
143 : /* No adaptive codebook (UC) */
144 8148 : set_zero( y1, L_SUBFR );
145 8148 : set_zero( exc + i_subfr, L_SUBFR );
146 :
147 : /*-----------------------------------------------------------------*
148 : * Gain clipping test to avoid unstable synthesis on frame erasure
149 : *-----------------------------------------------------------------*/
150 :
151 8148 : gp_clip( st->element_mode, st->core_brate, st->voicing, i_subfr, UNVOICED, xn, st->clip_var );
152 :
153 8148 : *pt_pitch = (float) L_SUBFR;
154 :
155 : /*-----------------------------------------------------------------*
156 : * Innovation encoding
157 : *-----------------------------------------------------------------*/
158 :
159 8148 : inov_encode( st, st->core_brate, 0, L_FRAME, st->last_L_frame, UNVOICED, st->bwidth, 1, i_subfr, -1, p_Aq, gain_pit, cn, exc, h1, hLPDmem->tilt_code, *pt_pitch, xn, code, y2, &unbits_PI, L_SUBFR );
160 :
161 8148 : E_corr_xy2( xn, y1, y2, g_corr2, L_SUBFR );
162 8148 : g_corr.y2y2 = 0.01F + g_corr2[2];
163 8148 : g_corr.xy2 = 0.01F + -0.5f * g_corr2[3];
164 8148 : g_corr.y1y2 = 0.01F + 0.5f * g_corr2[4];
165 :
166 8148 : g_corr.xx = 0.01F + dotp( xn, xn, L_SUBFR );
167 :
168 : /*----------------------------------------------------------------------*
169 : * Add Gaussian excitation
170 : *----------------------------------------------------------------------*/
171 :
172 8148 : assert( gain_pit == 0.f );
173 :
174 8148 : gauss_L2( h1, code2, y2, y22, &gain_code2, g_corr2, hLPDmem->tilt_code, p_Aq, FORMANT_SHARPENING_G1, &( st->seed_acelp ) );
175 :
176 8148 : g_corr.xy1 = 0.f;
177 8148 : g_corr.y1y1 = g_corr2[0];
178 8148 : g_corr.y1y2 = g_corr2[4];
179 :
180 : /*-----------------------------------------------------------------*
181 : * Gain encoding
182 : *-----------------------------------------------------------------*/
183 :
184 8148 : index = gain_enc_gacelp_uv( code, code2, L_SUBFR, Es_pred, &gain_pit, &gain_code, &gain_code2, &g_corr, &norm_gain_code, &gain_inov, st->flag_noisy_speech_snr );
185 :
186 : #ifdef DEBUGGING
187 : assert( st->acelp_cfg.gains_mode[i_subfr / L_SUBFR] == 7 && "Error: UC two-stage, only 5+2 gain Q is supported" );
188 : #endif
189 8148 : push_indice( st->hBstr, IND_GAIN, index, st->acelp_cfg.gains_mode[i_subfr / L_SUBFR] );
190 :
191 8148 : gp_clip_test_gain_pit( st->element_mode, st->core_brate, gain_pit, st->clip_var );
192 :
193 : /*----------------------------------------------------------*
194 : * - voice factor (for codebook tilt sharpening) *
195 : *----------------------------------------------------------*/
196 :
197 8148 : hLPDmem->tilt_code = est_tilt( exc + i_subfr, gain_pit, code, gain_code, &voice_fac, L_SUBFR, 0 );
198 :
199 : /*-----------------------------------------------------------------*
200 : * Update memory of the weighting filter
201 : *-----------------------------------------------------------------*/
202 :
203 8148 : hLPDmem->mem_w0 = xn[L_SUBFR - 1] - gain_code * y2[L_SUBFR - 1] - gain_code2 * y22[L_SUBFR - 1];
204 :
205 : /*-------------------------------------------------------*
206 : * - Find the total excitation. *
207 : *-------------------------------------------------------*/
208 :
209 529620 : for ( i = 0; i < L_SUBFR; i++ )
210 : {
211 521472 : exc2[i] = gain_pit * exc[i + i_subfr];
212 521472 : exc2[i] += gain_code2 * code2[i];
213 521472 : exc[i + i_subfr] = exc2[i] + gain_code * code[i];
214 : }
215 : }
216 :
217 8148 : *tmp_noise = norm_gain_code;
218 :
219 8148 : voice_factors[i_subfr / L_SUBFR] = 0.0f;
220 :
221 8148 : if ( st->hBWE_TD != NULL )
222 : {
223 8148 : interp_code_5over2( &exc[i_subfr], &bwe_exc[i_subfr * HIBND_ACB_L_FAC], L_SUBFR );
224 : }
225 :
226 : /*-----------------------------------------------------------------*
227 : * Synthesize speech to update mem_syn[].
228 : * Update A(z) filters
229 : *-----------------------------------------------------------------*/
230 :
231 8148 : syn_filt( p_Aq, M, &exc[i_subfr], &syn[i_subfr], L_SUBFR, hLPDmem->mem_syn, 1 );
232 :
233 8148 : p_Aw += ( M + 1 );
234 8148 : p_Aq += ( M + 1 );
235 8148 : pt_pitch++;
236 : }
237 :
238 : /* SC-VBR */
239 2037 : if ( st->Opt_SC_VBR )
240 : {
241 0 : st->hSC_VBR->prev_ppp_gain_pit = gain_pit;
242 0 : st->hSC_VBR->prev_tilt_code = hLPDmem->tilt_code;
243 : }
244 :
245 2037 : return;
246 : }
|
