|
8096
|
1 /*
|
|
|
2 * QCELP decoder
|
|
|
3 * Copyright (c) 2007 Reynaldo H. Verdejo Pinochet
|
|
|
4 *
|
|
|
5 * This file is part of FFmpeg.
|
|
|
6 *
|
|
|
7 * FFmpeg is free software; you can redistribute it and/or
|
|
|
8 * modify it under the terms of the GNU Lesser General Public
|
|
|
9 * License as published by the Free Software Foundation; either
|
|
|
10 * version 2.1 of the License, or (at your option) any later version.
|
|
|
11 *
|
|
|
12 * FFmpeg is distributed in the hope that it will be useful,
|
|
|
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
15 * Lesser General Public License for more details.
|
|
|
16 *
|
|
|
17 * You should have received a copy of the GNU Lesser General Public
|
|
|
18 * License along with FFmpeg; if not, write to the Free Software
|
|
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
20 */
|
|
8150
|
21
|
|
8096
|
22 /**
|
|
|
23 * @file qcelpdec.c
|
|
|
24 * QCELP decoder
|
|
|
25 * @author Reynaldo H. Verdejo Pinochet
|
|
|
26 */
|
|
|
27
|
|
|
28 #include <stddef.h>
|
|
|
29
|
|
|
30 #include "avcodec.h"
|
|
|
31 #include "bitstream.h"
|
|
|
32
|
|
|
33 #include "qcelp.h"
|
|
|
34 #include "qcelpdata.h"
|
|
|
35
|
|
|
36 #include "celp_math.h"
|
|
|
37 #include "celp_filters.h"
|
|
|
38
|
|
|
39 #undef NDEBUG
|
|
|
40 #include <assert.h>
|
|
|
41
|
|
8150
|
42 static void weighted_vector_sumf(float *out, const float *in_a,
|
|
|
43 const float *in_b, float weight_coeff_a,
|
|
|
44 float weight_coeff_b, int length)
|
|
|
45 {
|
|
|
46 int i;
|
|
8123
|
47
|
|
8150
|
48 for(i=0; i<length; i++)
|
|
8123
|
49 out[i] = weight_coeff_a * in_a[i]
|
|
|
50 + weight_coeff_b * in_b[i];
|
|
|
51 }
|
|
|
52
|
|
8096
|
53 /**
|
|
8145
|
54 * Initialize the speech codec according to the specification.
|
|
|
55 *
|
|
|
56 * TIA/EIA/IS-733 2.4.9
|
|
|
57 */
|
|
|
58 static av_cold int qcelp_decode_init(AVCodecContext *avctx) {
|
|
|
59 QCELPContext *q = avctx->priv_data;
|
|
|
60 int i;
|
|
|
61
|
|
|
62 avctx->sample_fmt = SAMPLE_FMT_FLT;
|
|
|
63
|
|
|
64 for (i = 0; i < 10; i++)
|
|
|
65 q->prev_lspf[i] = (i + 1) / 11.;
|
|
|
66
|
|
|
67 return 0;
|
|
|
68 }
|
|
|
69
|
|
|
70 /**
|
|
|
71 * Computes the scaled codebook vector Cdn From INDEX and GAIN
|
|
|
72 * for all rates.
|
|
|
73 *
|
|
|
74 * The specification lacks some information here.
|
|
|
75 *
|
|
|
76 * TIA/EIA/IS-733 has an omission on the codebook index determination
|
|
|
77 * formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says
|
|
|
78 * you have to subtract the decoded index parameter from the given scaled
|
|
|
79 * codebook vector index 'n' to get the desired circular codebook index, but
|
|
|
80 * it does not mention that you have to clamp 'n' to [0-9] in order to get
|
|
|
81 * RI-compliant results.
|
|
|
82 *
|
|
|
83 * The reason for this mistake seems to be the fact they forgot to mention you
|
|
|
84 * have to do these calculations per codebook subframe and adjust given
|
|
|
85 * equation values accordingly.
|
|
|
86 *
|
|
|
87 * @param q the context
|
|
|
88 * @param gain array holding the 4 pitch subframe gain values
|
|
|
89 * @param cdn_vector array for the generated scaled codebook vector
|
|
|
90 */
|
|
|
91 static void compute_svector(const QCELPContext *q,
|
|
|
92 const float *gain,
|
|
|
93 float *cdn_vector) {
|
|
|
94 int i, j, k;
|
|
|
95 uint16_t cbseed, cindex;
|
|
|
96 float *rnd, tmp_gain, fir_filter_value;
|
|
|
97
|
|
|
98 switch (q->framerate) {
|
|
|
99 case RATE_FULL:
|
|
|
100 for (i = 0; i < 16; i++) {
|
|
|
101 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
|
|
|
102 cindex = -q->cindex[i];
|
|
|
103 for (j = 0; j < 10; j++)
|
|
|
104 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127];
|
|
|
105 }
|
|
|
106 break;
|
|
|
107 case RATE_HALF:
|
|
|
108 for (i = 0; i < 4; i++) {
|
|
|
109 tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO;
|
|
|
110 cindex = -q->cindex[i];
|
|
|
111 for (j = 0; j < 40; j++)
|
|
|
112 *cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127];
|
|
|
113 }
|
|
|
114 break;
|
|
|
115 case RATE_QUARTER:
|
|
|
116 cbseed = (0x0003 & q->lspv[4])<<14 |
|
|
|
117 (0x003F & q->lspv[3])<< 8 |
|
|
|
118 (0x0060 & q->lspv[2])<< 1 |
|
|
|
119 (0x0007 & q->lspv[1])<< 3 |
|
|
|
120 (0x0038 & q->lspv[0])>> 3 ;
|
|
|
121 rnd = q->rnd_fir_filter_mem + 20;
|
|
|
122 for (i = 0; i < 8; i++) {
|
|
|
123 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
|
|
|
124 for (k = 0; k < 20; k++) {
|
|
|
125 cbseed = 521 * cbseed + 259;
|
|
|
126 *rnd = (int16_t)cbseed;
|
|
|
127
|
|
|
128 // FIR filter
|
|
|
129 fir_filter_value = 0.0;
|
|
|
130 for (j = 0; j < 10; j++)
|
|
|
131 fir_filter_value += qcelp_rnd_fir_coefs[j ] * (rnd[-j ] + rnd[-20+j]);
|
|
|
132 fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10];
|
|
|
133
|
|
|
134 *cdn_vector++ = tmp_gain * fir_filter_value;
|
|
|
135 rnd++;
|
|
|
136 }
|
|
|
137 }
|
|
|
138 memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float));
|
|
|
139 break;
|
|
|
140 case RATE_OCTAVE:
|
|
|
141 cbseed = q->first16bits;
|
|
|
142 for (i = 0; i < 8; i++) {
|
|
|
143 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0);
|
|
|
144 for (j = 0; j < 20; j++) {
|
|
|
145 cbseed = 521 * cbseed + 259;
|
|
|
146 *cdn_vector++ = tmp_gain * (int16_t)cbseed;
|
|
|
147 }
|
|
|
148 }
|
|
|
149 break;
|
|
|
150 case I_F_Q:
|
|
|
151 cbseed = -44; // random codebook index
|
|
|
152 for (i = 0; i < 4; i++) {
|
|
|
153 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO;
|
|
|
154 for (j = 0; j < 40; j++)
|
|
|
155 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127];
|
|
|
156 }
|
|
|
157 break;
|
|
|
158 }
|
|
|
159 }
|
|
|
160
|
|
|
161 /**
|
|
|
162 * Apply generic gain control.
|
|
|
163 *
|
|
|
164 * @param v_out output vector
|
|
|
165 * @param v_in gain-controlled vector
|
|
|
166 * @param v_ref vector to control gain of
|
|
|
167 *
|
|
|
168 * FIXME: If v_ref is a zero vector, it energy is zero
|
|
|
169 * and the behavior of the gain control is
|
|
|
170 * undefined in the specs.
|
|
|
171 *
|
|
|
172 * TIA/EIA/IS-733 2.4.8.3-2/3/4/5, 2.4.8.6
|
|
|
173 */
|
|
|
174 static void apply_gain_ctrl(float *v_out,
|
|
|
175 const float *v_ref,
|
|
|
176 const float *v_in) {
|
|
|
177 int i, j, len;
|
|
|
178 float scalefactor;
|
|
|
179
|
|
|
180 for (i = 0, j = 0; i < 4; i++) {
|
|
|
181 scalefactor = ff_dot_productf(v_in + j, v_in + j, 40);
|
|
|
182 if (scalefactor)
|
|
|
183 scalefactor = sqrt(ff_dot_productf(v_ref + j, v_ref + j, 40) / scalefactor);
|
|
|
184 else
|
|
|
185 av_log_missing_feature(NULL, "Zero energy for gain control", 1);
|
|
|
186 for (len = j + 40; j < len; j++)
|
|
|
187 v_out[j] = scalefactor * v_in[j];
|
|
|
188 }
|
|
|
189 }
|
|
|
190
|
|
|
191 /**
|
|
8096
|
192 * Apply filter in pitch-subframe steps.
|
|
|
193 *
|
|
|
194 * @param memory buffer for the previous state of the filter
|
|
|
195 * - must be able to contain 303 elements
|
|
|
196 * - the 143 first elements are from the previous state
|
|
|
197 * - the next 160 are for output
|
|
|
198 * @param v_in input filter vector
|
|
|
199 * @param gain per-subframe gain array, each element is between 0.0 and 2.0
|
|
|
200 * @param lag per-subframe lag array, each element is
|
|
|
201 * - between 16 and 143 if its corresponding pfrac is 0,
|
|
|
202 * - between 16 and 139 otherwise
|
|
|
203 * @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0 otherwise
|
|
|
204 *
|
|
|
205 * @return filter output vector
|
|
|
206 */
|
|
8150
|
207 static const float *do_pitchfilter(float memory[303], const float v_in[160],
|
|
|
208 const float gain[4], const uint8_t *lag,
|
|
|
209 const uint8_t pfrac[4])
|
|
|
210 {
|
|
8096
|
211 int i, j;
|
|
|
212 float *v_lag, *v_out;
|
|
|
213 const float *v_len;
|
|
|
214
|
|
|
215 v_out = memory + 143; // Output vector starts at memory[143].
|
|
|
216
|
|
8150
|
217 for(i=0; i<4; i++)
|
|
|
218 {
|
|
|
219 if(gain[i])
|
|
|
220 {
|
|
8096
|
221 v_lag = memory + 143 + 40 * i - lag[i];
|
|
8150
|
222 for(v_len=v_in+40; v_in<v_len; v_in++)
|
|
|
223 {
|
|
|
224 if(pfrac[i]) // If it is a fractional lag...
|
|
|
225 {
|
|
|
226 for(j=0, *v_out=0.; j<4; j++)
|
|
8096
|
227 *v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]);
|
|
8150
|
228 }else
|
|
8096
|
229 *v_out = *v_lag;
|
|
|
230
|
|
|
231 *v_out = *v_in + gain[i] * *v_out;
|
|
|
232
|
|
|
233 v_lag++;
|
|
|
234 v_out++;
|
|
|
235 }
|
|
8150
|
236 }else
|
|
|
237 {
|
|
8096
|
238 memcpy(v_out, v_in, 40 * sizeof(float));
|
|
|
239 v_in += 40;
|
|
|
240 v_out += 40;
|
|
|
241 }
|
|
8150
|
242 }
|
|
8096
|
243
|
|
|
244 memmove(memory, memory + 160, 143 * sizeof(float));
|
|
|
245 return memory + 143;
|
|
|
246 }
|
|
|
247
|
|
8127
|
248 /**
|
|
|
249 * Interpolates LSP frequencies and computes LPC coefficients
|
|
|
250 * for a given framerate & pitch subframe.
|
|
|
251 *
|
|
|
252 * TIA/EIA/IS-733 2.4.3.3.4
|
|
|
253 *
|
|
|
254 * @param q the context
|
|
|
255 * @param curr_lspf LSP frequencies vector of the current frame
|
|
|
256 * @param lpc float vector for the resulting LPC
|
|
|
257 * @param subframe_num frame number in decoded stream
|
|
|
258 */
|
|
8150
|
259 void interpolate_lpc(QCELPContext *q, const float *curr_lspf, float *lpc,
|
|
|
260 const int subframe_num)
|
|
|
261 {
|
|
8127
|
262 float interpolated_lspf[10];
|
|
|
263 float weight;
|
|
|
264
|
|
8150
|
265 if(q->framerate >= RATE_QUARTER)
|
|
8127
|
266 weight = 0.25 * (subframe_num + 1);
|
|
8150
|
267 else if(q->framerate == RATE_OCTAVE && !subframe_num)
|
|
8127
|
268 weight = 0.625;
|
|
8150
|
269 else
|
|
8127
|
270 weight = 1.0;
|
|
|
271
|
|
8150
|
272 if(weight != 1.0)
|
|
|
273 {
|
|
|
274 weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf,
|
|
|
275 weight, 1.0 - weight, 10);
|
|
8145
|
276 qcelp_lspf2lpc(interpolated_lspf, lpc);
|
|
8150
|
277 }else if(q->framerate >= RATE_QUARTER || (q->framerate == I_F_Q && !subframe_num))
|
|
8145
|
278 qcelp_lspf2lpc(curr_lspf, lpc);
|
|
8127
|
279 }
|
|
|
280
|
|
8150
|
281 static int buf_size2framerate(const int buf_size)
|
|
|
282 {
|
|
|
283 switch(buf_size)
|
|
|
284 {
|
|
|
285 case 35:
|
|
|
286 return RATE_FULL;
|
|
|
287 case 17:
|
|
|
288 return RATE_HALF;
|
|
|
289 case 8:
|
|
|
290 return RATE_QUARTER;
|
|
|
291 case 4:
|
|
|
292 return RATE_OCTAVE;
|
|
|
293 case 1:
|
|
|
294 return SILENCE;
|
|
8123
|
295 }
|
|
8150
|
296
|
|
8123
|
297 return -1;
|
|
|
298 }
|
|
|
299
|
|
8096
|
300 static void warn_insufficient_frame_quality(AVCodecContext *avctx,
|
|
8150
|
301 const char *message)
|
|
|
302 {
|
|
|
303 av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number,
|
|
|
304 message);
|
|
8096
|
305 }
|
|
8127
|
306
|
|
|
307 AVCodec qcelp_decoder =
|
|
|
308 {
|
|
|
309 .name = "qcelp",
|
|
|
310 .type = CODEC_TYPE_AUDIO,
|
|
|
311 .id = CODEC_ID_QCELP,
|
|
|
312 .init = qcelp_decode_init,
|
|
|
313 .decode = qcelp_decode_frame,
|
|
|
314 .priv_data_size = sizeof(QCELPContext),
|
|
|
315 .long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"),
|
|
|
316 };
|