Mercurial > libavcodec.hg
annotate qcelpdec.c @ 8151:e2c068cb210a libavcodec
Credit Kenan Gillet for his contributions towards merging the SoC QCELP decoder.
author | reynaldo |
---|---|
date | Sun, 16 Nov 2008 01:00:25 +0000 |
parents | 4da8fc62ae00 |
children | cc1e8c59f1e8 |
rev | line source |
---|---|
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 | |
8151
e2c068cb210a
Credit Kenan Gillet for his contributions towards merging the SoC QCELP decoder.
reynaldo
parents:
8150
diff
changeset
|
26 * @remark FFmpeg merging spearheaded by Kenan Gillet |
8096 | 27 */ |
28 | |
29 #include <stddef.h> | |
30 | |
31 #include "avcodec.h" | |
32 #include "bitstream.h" | |
33 | |
34 #include "qcelp.h" | |
35 #include "qcelpdata.h" | |
36 | |
37 #include "celp_math.h" | |
38 #include "celp_filters.h" | |
39 | |
40 #undef NDEBUG | |
41 #include <assert.h> | |
42 | |
8150 | 43 static void weighted_vector_sumf(float *out, const float *in_a, |
44 const float *in_b, float weight_coeff_a, | |
45 float weight_coeff_b, int length) | |
46 { | |
47 int i; | |
8123 | 48 |
8150 | 49 for(i=0; i<length; i++) |
8123 | 50 out[i] = weight_coeff_a * in_a[i] |
51 + weight_coeff_b * in_b[i]; | |
52 } | |
53 | |
8096 | 54 /** |
8145 | 55 * Initialize the speech codec according to the specification. |
56 * | |
57 * TIA/EIA/IS-733 2.4.9 | |
58 */ | |
59 static av_cold int qcelp_decode_init(AVCodecContext *avctx) { | |
60 QCELPContext *q = avctx->priv_data; | |
61 int i; | |
62 | |
63 avctx->sample_fmt = SAMPLE_FMT_FLT; | |
64 | |
65 for (i = 0; i < 10; i++) | |
66 q->prev_lspf[i] = (i + 1) / 11.; | |
67 | |
68 return 0; | |
69 } | |
70 | |
71 /** | |
72 * Computes the scaled codebook vector Cdn From INDEX and GAIN | |
73 * for all rates. | |
74 * | |
75 * The specification lacks some information here. | |
76 * | |
77 * TIA/EIA/IS-733 has an omission on the codebook index determination | |
78 * formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says | |
79 * you have to subtract the decoded index parameter from the given scaled | |
80 * codebook vector index 'n' to get the desired circular codebook index, but | |
81 * it does not mention that you have to clamp 'n' to [0-9] in order to get | |
82 * RI-compliant results. | |
83 * | |
84 * The reason for this mistake seems to be the fact they forgot to mention you | |
85 * have to do these calculations per codebook subframe and adjust given | |
86 * equation values accordingly. | |
87 * | |
88 * @param q the context | |
89 * @param gain array holding the 4 pitch subframe gain values | |
90 * @param cdn_vector array for the generated scaled codebook vector | |
91 */ | |
92 static void compute_svector(const QCELPContext *q, | |
93 const float *gain, | |
94 float *cdn_vector) { | |
95 int i, j, k; | |
96 uint16_t cbseed, cindex; | |
97 float *rnd, tmp_gain, fir_filter_value; | |
98 | |
99 switch (q->framerate) { | |
100 case RATE_FULL: | |
101 for (i = 0; i < 16; i++) { | |
102 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; | |
103 cindex = -q->cindex[i]; | |
104 for (j = 0; j < 10; j++) | |
105 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127]; | |
106 } | |
107 break; | |
108 case RATE_HALF: | |
109 for (i = 0; i < 4; i++) { | |
110 tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO; | |
111 cindex = -q->cindex[i]; | |
112 for (j = 0; j < 40; j++) | |
113 *cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127]; | |
114 } | |
115 break; | |
116 case RATE_QUARTER: | |
117 cbseed = (0x0003 & q->lspv[4])<<14 | | |
118 (0x003F & q->lspv[3])<< 8 | | |
119 (0x0060 & q->lspv[2])<< 1 | | |
120 (0x0007 & q->lspv[1])<< 3 | | |
121 (0x0038 & q->lspv[0])>> 3 ; | |
122 rnd = q->rnd_fir_filter_mem + 20; | |
123 for (i = 0; i < 8; i++) { | |
124 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); | |
125 for (k = 0; k < 20; k++) { | |
126 cbseed = 521 * cbseed + 259; | |
127 *rnd = (int16_t)cbseed; | |
128 | |
129 // FIR filter | |
130 fir_filter_value = 0.0; | |
131 for (j = 0; j < 10; j++) | |
132 fir_filter_value += qcelp_rnd_fir_coefs[j ] * (rnd[-j ] + rnd[-20+j]); | |
133 fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10]; | |
134 | |
135 *cdn_vector++ = tmp_gain * fir_filter_value; | |
136 rnd++; | |
137 } | |
138 } | |
139 memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float)); | |
140 break; | |
141 case RATE_OCTAVE: | |
142 cbseed = q->first16bits; | |
143 for (i = 0; i < 8; i++) { | |
144 tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); | |
145 for (j = 0; j < 20; j++) { | |
146 cbseed = 521 * cbseed + 259; | |
147 *cdn_vector++ = tmp_gain * (int16_t)cbseed; | |
148 } | |
149 } | |
150 break; | |
151 case I_F_Q: | |
152 cbseed = -44; // random codebook index | |
153 for (i = 0; i < 4; i++) { | |
154 tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; | |
155 for (j = 0; j < 40; j++) | |
156 *cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127]; | |
157 } | |
158 break; | |
159 } | |
160 } | |
161 | |
162 /** | |
163 * Apply generic gain control. | |
164 * | |
165 * @param v_out output vector | |
166 * @param v_in gain-controlled vector | |
167 * @param v_ref vector to control gain of | |
168 * | |
169 * FIXME: If v_ref is a zero vector, it energy is zero | |
170 * and the behavior of the gain control is | |
171 * undefined in the specs. | |
172 * | |
173 * TIA/EIA/IS-733 2.4.8.3-2/3/4/5, 2.4.8.6 | |
174 */ | |
175 static void apply_gain_ctrl(float *v_out, | |
176 const float *v_ref, | |
177 const float *v_in) { | |
178 int i, j, len; | |
179 float scalefactor; | |
180 | |
181 for (i = 0, j = 0; i < 4; i++) { | |
182 scalefactor = ff_dot_productf(v_in + j, v_in + j, 40); | |
183 if (scalefactor) | |
184 scalefactor = sqrt(ff_dot_productf(v_ref + j, v_ref + j, 40) / scalefactor); | |
185 else | |
186 av_log_missing_feature(NULL, "Zero energy for gain control", 1); | |
187 for (len = j + 40; j < len; j++) | |
188 v_out[j] = scalefactor * v_in[j]; | |
189 } | |
190 } | |
191 | |
192 /** | |
8096 | 193 * Apply filter in pitch-subframe steps. |
194 * | |
195 * @param memory buffer for the previous state of the filter | |
196 * - must be able to contain 303 elements | |
197 * - the 143 first elements are from the previous state | |
198 * - the next 160 are for output | |
199 * @param v_in input filter vector | |
200 * @param gain per-subframe gain array, each element is between 0.0 and 2.0 | |
201 * @param lag per-subframe lag array, each element is | |
202 * - between 16 and 143 if its corresponding pfrac is 0, | |
203 * - between 16 and 139 otherwise | |
204 * @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0 otherwise | |
205 * | |
206 * @return filter output vector | |
207 */ | |
8150 | 208 static const float *do_pitchfilter(float memory[303], const float v_in[160], |
209 const float gain[4], const uint8_t *lag, | |
210 const uint8_t pfrac[4]) | |
211 { | |
8096 | 212 int i, j; |
213 float *v_lag, *v_out; | |
214 const float *v_len; | |
215 | |
216 v_out = memory + 143; // Output vector starts at memory[143]. | |
217 | |
8150 | 218 for(i=0; i<4; i++) |
219 { | |
220 if(gain[i]) | |
221 { | |
8096 | 222 v_lag = memory + 143 + 40 * i - lag[i]; |
8150 | 223 for(v_len=v_in+40; v_in<v_len; v_in++) |
224 { | |
225 if(pfrac[i]) // If it is a fractional lag... | |
226 { | |
227 for(j=0, *v_out=0.; j<4; j++) | |
8096 | 228 *v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]); |
8150 | 229 }else |
8096 | 230 *v_out = *v_lag; |
231 | |
232 *v_out = *v_in + gain[i] * *v_out; | |
233 | |
234 v_lag++; | |
235 v_out++; | |
236 } | |
8150 | 237 }else |
238 { | |
8096 | 239 memcpy(v_out, v_in, 40 * sizeof(float)); |
240 v_in += 40; | |
241 v_out += 40; | |
242 } | |
8150 | 243 } |
8096 | 244 |
245 memmove(memory, memory + 160, 143 * sizeof(float)); | |
246 return memory + 143; | |
247 } | |
248 | |
8127 | 249 /** |
250 * Interpolates LSP frequencies and computes LPC coefficients | |
251 * for a given framerate & pitch subframe. | |
252 * | |
253 * TIA/EIA/IS-733 2.4.3.3.4 | |
254 * | |
255 * @param q the context | |
256 * @param curr_lspf LSP frequencies vector of the current frame | |
257 * @param lpc float vector for the resulting LPC | |
258 * @param subframe_num frame number in decoded stream | |
259 */ | |
8150 | 260 void interpolate_lpc(QCELPContext *q, const float *curr_lspf, float *lpc, |
261 const int subframe_num) | |
262 { | |
8127 | 263 float interpolated_lspf[10]; |
264 float weight; | |
265 | |
8150 | 266 if(q->framerate >= RATE_QUARTER) |
8127 | 267 weight = 0.25 * (subframe_num + 1); |
8150 | 268 else if(q->framerate == RATE_OCTAVE && !subframe_num) |
8127 | 269 weight = 0.625; |
8150 | 270 else |
8127 | 271 weight = 1.0; |
272 | |
8150 | 273 if(weight != 1.0) |
274 { | |
275 weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf, | |
276 weight, 1.0 - weight, 10); | |
8145 | 277 qcelp_lspf2lpc(interpolated_lspf, lpc); |
8150 | 278 }else if(q->framerate >= RATE_QUARTER || (q->framerate == I_F_Q && !subframe_num)) |
8145 | 279 qcelp_lspf2lpc(curr_lspf, lpc); |
8127 | 280 } |
281 | |
8150 | 282 static int buf_size2framerate(const int buf_size) |
283 { | |
284 switch(buf_size) | |
285 { | |
286 case 35: | |
287 return RATE_FULL; | |
288 case 17: | |
289 return RATE_HALF; | |
290 case 8: | |
291 return RATE_QUARTER; | |
292 case 4: | |
293 return RATE_OCTAVE; | |
294 case 1: | |
295 return SILENCE; | |
8123 | 296 } |
8150 | 297 |
8123 | 298 return -1; |
299 } | |
300 | |
8096 | 301 static void warn_insufficient_frame_quality(AVCodecContext *avctx, |
8150 | 302 const char *message) |
303 { | |
304 av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number, | |
305 message); | |
8096 | 306 } |
8127 | 307 |
308 AVCodec qcelp_decoder = | |
309 { | |
310 .name = "qcelp", | |
311 .type = CODEC_TYPE_AUDIO, | |
312 .id = CODEC_ID_QCELP, | |
313 .init = qcelp_decode_init, | |
314 .decode = qcelp_decode_frame, | |
315 .priv_data_size = sizeof(QCELPContext), | |
316 .long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"), | |
317 }; |