blob: 9c8f684003c588b2313fe8fe40ea431e79a7213f
1 | /* |
2 | * SIPR decoder for the 16k mode |
3 | * |
4 | * Copyright (c) 2008 Vladimir Voroshilov |
5 | * Copyright (c) 2009 Vitor Sessak |
6 | * |
7 | * This file is part of FFmpeg. |
8 | * |
9 | * FFmpeg is free software; you can redistribute it and/or |
10 | * modify it under the terms of the GNU Lesser General Public |
11 | * License as published by the Free Software Foundation; either |
12 | * version 2.1 of the License, or (at your option) any later version. |
13 | * |
14 | * FFmpeg is distributed in the hope that it will be useful, |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
17 | * Lesser General Public License for more details. |
18 | * |
19 | * You should have received a copy of the GNU Lesser General Public |
20 | * License along with FFmpeg; if not, write to the Free Software |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
22 | */ |
23 | |
24 | #include <math.h> |
25 | |
26 | #include "sipr.h" |
27 | #include "libavutil/attributes.h" |
28 | #include "libavutil/common.h" |
29 | #include "libavutil/float_dsp.h" |
30 | #include "libavutil/mathematics.h" |
31 | #include "lsp.h" |
32 | #include "acelp_vectors.h" |
33 | #include "acelp_pitch_delay.h" |
34 | #include "acelp_filters.h" |
35 | #include "celp_filters.h" |
36 | |
37 | #include "sipr16kdata.h" |
38 | |
39 | /** |
40 | * Convert an lsf vector into an lsp vector. |
41 | * |
42 | * @param lsf input lsf vector |
43 | * @param lsp output lsp vector |
44 | */ |
45 | static void lsf2lsp(const float *lsf, double *lsp) |
46 | { |
47 | int i; |
48 | |
49 | for (i = 0; i < LP_FILTER_ORDER_16k; i++) |
50 | lsp[i] = cosf(lsf[i]); |
51 | } |
52 | |
53 | static void dequant(float *out, const int *idx, const float * const cbs[]) |
54 | { |
55 | int i; |
56 | |
57 | for (i = 0; i < 4; i++) |
58 | memcpy(out + 3*i, cbs[i] + 3*idx[i], 3*sizeof(float)); |
59 | |
60 | memcpy(out + 12, cbs[4] + 4*idx[4], 4*sizeof(float)); |
61 | } |
62 | |
63 | static void lsf_decode_fp_16k(float* lsf_history, float* isp_new, |
64 | const int* parm, int ma_pred) |
65 | { |
66 | int i; |
67 | float isp_q[LP_FILTER_ORDER_16k]; |
68 | |
69 | dequant(isp_q, parm, lsf_codebooks_16k); |
70 | |
71 | for (i = 0; i < LP_FILTER_ORDER_16k; i++) { |
72 | isp_new[i] = (1 - qu[ma_pred]) * isp_q[i] |
73 | + qu[ma_pred] * lsf_history[i] |
74 | + mean_lsf_16k[i]; |
75 | } |
76 | |
77 | memcpy(lsf_history, isp_q, LP_FILTER_ORDER_16k * sizeof(float)); |
78 | } |
79 | |
80 | static int dec_delay3_1st(int index) |
81 | { |
82 | if (index < 390) { |
83 | return index + 88; |
84 | } else |
85 | return 3 * index - 690; |
86 | } |
87 | |
88 | static int dec_delay3_2nd(int index, int pit_min, int pit_max, |
89 | int pitch_lag_prev) |
90 | { |
91 | if (index < 62) { |
92 | int pitch_delay_min = av_clip(pitch_lag_prev - 10, |
93 | pit_min, pit_max - 19); |
94 | return 3 * pitch_delay_min + index - 2; |
95 | } else |
96 | return 3 * pitch_lag_prev; |
97 | } |
98 | |
99 | static void postfilter(float *out_data, float* synth, float* iir_mem, |
100 | float* filt_mem[2], float* mem_preemph) |
101 | { |
102 | float buf[30 + LP_FILTER_ORDER_16k]; |
103 | float *tmpbuf = buf + LP_FILTER_ORDER_16k; |
104 | float s; |
105 | int i; |
106 | |
107 | for (i = 0; i < LP_FILTER_ORDER_16k; i++) |
108 | filt_mem[0][i] = iir_mem[i] * ff_pow_0_5[i]; |
109 | |
110 | memcpy(tmpbuf - LP_FILTER_ORDER_16k, mem_preemph, |
111 | LP_FILTER_ORDER_16k*sizeof(*buf)); |
112 | |
113 | ff_celp_lp_synthesis_filterf(tmpbuf, filt_mem[1], synth, 30, |
114 | LP_FILTER_ORDER_16k); |
115 | |
116 | memcpy(synth - LP_FILTER_ORDER_16k, mem_preemph, |
117 | LP_FILTER_ORDER_16k * sizeof(*synth)); |
118 | |
119 | ff_celp_lp_synthesis_filterf(synth, filt_mem[0], synth, 30, |
120 | LP_FILTER_ORDER_16k); |
121 | |
122 | memcpy(out_data + 30 - LP_FILTER_ORDER_16k, |
123 | synth + 30 - LP_FILTER_ORDER_16k, |
124 | LP_FILTER_ORDER_16k * sizeof(*synth)); |
125 | |
126 | ff_celp_lp_synthesis_filterf(out_data + 30, filt_mem[0], |
127 | synth + 30, 2 * L_SUBFR_16k - 30, |
128 | LP_FILTER_ORDER_16k); |
129 | |
130 | |
131 | memcpy(mem_preemph, out_data + 2*L_SUBFR_16k - LP_FILTER_ORDER_16k, |
132 | LP_FILTER_ORDER_16k * sizeof(*synth)); |
133 | |
134 | FFSWAP(float *, filt_mem[0], filt_mem[1]); |
135 | for (i = 0, s = 0; i < 30; i++, s += 1.0/30) |
136 | out_data[i] = tmpbuf[i] + s * (synth[i] - tmpbuf[i]); |
137 | } |
138 | |
139 | /** |
140 | * Floating point version of ff_acelp_lp_decode(). |
141 | */ |
142 | static void acelp_lp_decodef(float *lp_1st, float *lp_2nd, |
143 | const double *lsp_2nd, const double *lsp_prev) |
144 | { |
145 | double lsp_1st[LP_FILTER_ORDER_16k]; |
146 | int i; |
147 | |
148 | /* LSP values for first subframe (3.2.5 of G.729, Equation 24) */ |
149 | for (i = 0; i < LP_FILTER_ORDER_16k; i++) |
150 | lsp_1st[i] = (lsp_2nd[i] + lsp_prev[i]) * 0.5; |
151 | |
152 | ff_acelp_lspd2lpc(lsp_1st, lp_1st, LP_FILTER_ORDER_16k >> 1); |
153 | |
154 | /* LSP values for second subframe (3.2.5 of G.729) */ |
155 | ff_acelp_lspd2lpc(lsp_2nd, lp_2nd, LP_FILTER_ORDER_16k >> 1); |
156 | } |
157 | |
158 | /** |
159 | * Floating point version of ff_acelp_decode_gain_code(). |
160 | */ |
161 | static float acelp_decode_gain_codef(float gain_corr_factor, const float *fc_v, |
162 | float mr_energy, const float *quant_energy, |
163 | const float *ma_prediction_coeff, |
164 | int subframe_size, int ma_pred_order) |
165 | { |
166 | mr_energy += avpriv_scalarproduct_float_c(quant_energy, ma_prediction_coeff, |
167 | ma_pred_order); |
168 | |
169 | mr_energy = gain_corr_factor * exp(M_LN10 / 20. * mr_energy) / |
170 | sqrt((0.01 + avpriv_scalarproduct_float_c(fc_v, fc_v, subframe_size))); |
171 | return mr_energy; |
172 | } |
173 | |
174 | #define DIVIDE_BY_3(x) ((x) * 10923 >> 15) |
175 | |
176 | void ff_sipr_decode_frame_16k(SiprContext *ctx, SiprParameters *params, |
177 | float *out_data) |
178 | { |
179 | int frame_size = SUBFRAME_COUNT_16k * L_SUBFR_16k; |
180 | float *synth = ctx->synth_buf + LP_FILTER_ORDER_16k; |
181 | float lsf_new[LP_FILTER_ORDER_16k]; |
182 | double lsp_new[LP_FILTER_ORDER_16k]; |
183 | float Az[2][LP_FILTER_ORDER_16k]; |
184 | float fixed_vector[L_SUBFR_16k]; |
185 | float pitch_fac, gain_code; |
186 | |
187 | int i; |
188 | int pitch_delay_3x; |
189 | |
190 | float *excitation = ctx->excitation + 292; |
191 | |
192 | lsf_decode_fp_16k(ctx->lsf_history, lsf_new, params->vq_indexes, |
193 | params->ma_pred_switch); |
194 | |
195 | ff_set_min_dist_lsf(lsf_new, LSFQ_DIFF_MIN / 2, LP_FILTER_ORDER_16k); |
196 | |
197 | lsf2lsp(lsf_new, lsp_new); |
198 | |
199 | acelp_lp_decodef(Az[0], Az[1], lsp_new, ctx->lsp_history_16k); |
200 | |
201 | memcpy(ctx->lsp_history_16k, lsp_new, LP_FILTER_ORDER_16k * sizeof(double)); |
202 | |
203 | memcpy(synth - LP_FILTER_ORDER_16k, ctx->synth, |
204 | LP_FILTER_ORDER_16k * sizeof(*synth)); |
205 | |
206 | for (i = 0; i < SUBFRAME_COUNT_16k; i++) { |
207 | int i_subfr = i * L_SUBFR_16k; |
208 | AMRFixed f; |
209 | float gain_corr_factor; |
210 | int pitch_delay_int; |
211 | int pitch_delay_frac; |
212 | |
213 | if (!i) { |
214 | pitch_delay_3x = dec_delay3_1st(params->pitch_delay[i]); |
215 | } else |
216 | pitch_delay_3x = dec_delay3_2nd(params->pitch_delay[i], |
217 | PITCH_MIN, PITCH_MAX, |
218 | ctx->pitch_lag_prev); |
219 | |
220 | pitch_fac = gain_pitch_cb_16k[params->gp_index[i]]; |
221 | f.pitch_fac = FFMIN(pitch_fac, 1.0); |
222 | f.pitch_lag = DIVIDE_BY_3(pitch_delay_3x+1); |
223 | ctx->pitch_lag_prev = f.pitch_lag; |
224 | |
225 | pitch_delay_int = DIVIDE_BY_3(pitch_delay_3x + 2); |
226 | pitch_delay_frac = pitch_delay_3x + 2 - 3*pitch_delay_int; |
227 | |
228 | ff_acelp_interpolatef(&excitation[i_subfr], |
229 | &excitation[i_subfr] - pitch_delay_int + 1, |
230 | sinc_win, 3, pitch_delay_frac + 1, |
231 | LP_FILTER_ORDER, L_SUBFR_16k); |
232 | |
233 | |
234 | memset(fixed_vector, 0, sizeof(fixed_vector)); |
235 | |
236 | ff_decode_10_pulses_35bits(params->fc_indexes[i], &f, |
237 | ff_fc_4pulses_8bits_tracks_13, 5, 4); |
238 | |
239 | ff_set_fixed_vector(fixed_vector, &f, 1.0, L_SUBFR_16k); |
240 | |
241 | gain_corr_factor = gain_cb_16k[params->gc_index[i]]; |
242 | gain_code = gain_corr_factor * |
243 | acelp_decode_gain_codef(sqrt(L_SUBFR_16k), fixed_vector, |
244 | 19.0 - 15.0/(0.05*M_LN10/M_LN2), |
245 | pred_16k, ctx->energy_history, |
246 | L_SUBFR_16k, 2); |
247 | |
248 | ctx->energy_history[1] = ctx->energy_history[0]; |
249 | ctx->energy_history[0] = 20.0 * log10f(gain_corr_factor); |
250 | |
251 | ff_weighted_vector_sumf(&excitation[i_subfr], &excitation[i_subfr], |
252 | fixed_vector, pitch_fac, |
253 | gain_code, L_SUBFR_16k); |
254 | |
255 | ff_celp_lp_synthesis_filterf(synth + i_subfr, Az[i], |
256 | &excitation[i_subfr], L_SUBFR_16k, |
257 | LP_FILTER_ORDER_16k); |
258 | |
259 | } |
260 | memcpy(ctx->synth, synth + frame_size - LP_FILTER_ORDER_16k, |
261 | LP_FILTER_ORDER_16k * sizeof(*synth)); |
262 | |
263 | memmove(ctx->excitation, ctx->excitation + 2 * L_SUBFR_16k, |
264 | (L_INTERPOL+PITCH_MAX) * sizeof(float)); |
265 | |
266 | postfilter(out_data, synth, ctx->iir_mem, ctx->filt_mem, ctx->mem_preemph); |
267 | |
268 | memcpy(ctx->iir_mem, Az[1], LP_FILTER_ORDER_16k * sizeof(float)); |
269 | } |
270 | |
271 | av_cold void ff_sipr_init_16k(SiprContext *ctx) |
272 | { |
273 | int i; |
274 | |
275 | for (i = 0; i < LP_FILTER_ORDER_16k; i++) |
276 | ctx->lsp_history_16k[i] = cos((i + 1) * M_PI/(LP_FILTER_ORDER_16k + 1)); |
277 | |
278 | ctx->filt_mem[0] = ctx->filt_buf[0]; |
279 | ctx->filt_mem[1] = ctx->filt_buf[1]; |
280 | |
281 | ctx->pitch_lag_prev = 180; |
282 | } |
283 |