ffmpeg / libavcodec / qcelpdec.c @ d36beb3f
History | View | Annotate | Download (25.9 KB)
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 |
*/
|
21 |
|
22 |
/**
|
23 |
* @file
|
24 |
* QCELP decoder
|
25 |
* @author Reynaldo H. Verdejo Pinochet
|
26 |
* @remark FFmpeg merging spearheaded by Kenan Gillet
|
27 |
* @remark Development mentored by Benjamin Larson
|
28 |
*/
|
29 |
|
30 |
#include <stddef.h> |
31 |
|
32 |
#include "avcodec.h" |
33 |
#include "internal.h" |
34 |
#include "get_bits.h" |
35 |
|
36 |
#include "qcelpdata.h" |
37 |
|
38 |
#include "celp_math.h" |
39 |
#include "celp_filters.h" |
40 |
#include "acelp_filters.h" |
41 |
#include "acelp_vectors.h" |
42 |
#include "lsp.h" |
43 |
|
44 |
#undef NDEBUG
|
45 |
#include <assert.h> |
46 |
|
47 |
typedef enum |
48 |
{ |
49 |
I_F_Q = -1, /*!< insufficient frame quality */ |
50 |
SILENCE, |
51 |
RATE_OCTAVE, |
52 |
RATE_QUARTER, |
53 |
RATE_HALF, |
54 |
RATE_FULL |
55 |
} qcelp_packet_rate; |
56 |
|
57 |
typedef struct |
58 |
{ |
59 |
GetBitContext gb; |
60 |
qcelp_packet_rate bitrate; |
61 |
QCELPFrame frame; /*!< unpacked data frame */
|
62 |
|
63 |
uint8_t erasure_count; |
64 |
uint8_t octave_count; /*!< count the consecutive RATE_OCTAVE frames */
|
65 |
float prev_lspf[10]; |
66 |
float predictor_lspf[10];/*!< LSP predictor for RATE_OCTAVE and I_F_Q */ |
67 |
float pitch_synthesis_filter_mem[303]; |
68 |
float pitch_pre_filter_mem[303]; |
69 |
float rnd_fir_filter_mem[180]; |
70 |
float formant_mem[170]; |
71 |
float last_codebook_gain;
|
72 |
int prev_g1[2]; |
73 |
int prev_bitrate;
|
74 |
float pitch_gain[4]; |
75 |
uint8_t pitch_lag[4];
|
76 |
uint16_t first16bits; |
77 |
uint8_t warned_buf_mismatch_bitrate; |
78 |
|
79 |
/* postfilter */
|
80 |
float postfilter_synth_mem[10]; |
81 |
float postfilter_agc_mem;
|
82 |
float postfilter_tilt_mem;
|
83 |
} QCELPContext; |
84 |
|
85 |
/**
|
86 |
* Initialize the speech codec according to the specification.
|
87 |
*
|
88 |
* TIA/EIA/IS-733 2.4.9
|
89 |
*/
|
90 |
static av_cold int qcelp_decode_init(AVCodecContext *avctx) |
91 |
{ |
92 |
QCELPContext *q = avctx->priv_data; |
93 |
int i;
|
94 |
|
95 |
avctx->sample_fmt = AV_SAMPLE_FMT_FLT; |
96 |
|
97 |
for(i=0; i<10; i++) |
98 |
q->prev_lspf[i] = (i+1)/11.; |
99 |
|
100 |
return 0; |
101 |
} |
102 |
|
103 |
/**
|
104 |
* Decode the 10 quantized LSP frequencies from the LSPV/LSP
|
105 |
* transmission codes of any bitrate and check for badly received packets.
|
106 |
*
|
107 |
* @param q the context
|
108 |
* @param lspf line spectral pair frequencies
|
109 |
*
|
110 |
* @return 0 on success, -1 if the packet is badly received
|
111 |
*
|
112 |
* TIA/EIA/IS-733 2.4.3.2.6.2-2, 2.4.8.7.3
|
113 |
*/
|
114 |
static int decode_lspf(QCELPContext *q, float *lspf) |
115 |
{ |
116 |
int i;
|
117 |
float tmp_lspf, smooth, erasure_coeff;
|
118 |
const float *predictors; |
119 |
|
120 |
if(q->bitrate == RATE_OCTAVE || q->bitrate == I_F_Q)
|
121 |
{ |
122 |
predictors = (q->prev_bitrate != RATE_OCTAVE && |
123 |
q->prev_bitrate != I_F_Q ? |
124 |
q->prev_lspf : q->predictor_lspf); |
125 |
|
126 |
if(q->bitrate == RATE_OCTAVE)
|
127 |
{ |
128 |
q->octave_count++; |
129 |
|
130 |
for(i=0; i<10; i++) |
131 |
{ |
132 |
q->predictor_lspf[i] = |
133 |
lspf[i] = (q->frame.lspv[i] ? QCELP_LSP_SPREAD_FACTOR |
134 |
: -QCELP_LSP_SPREAD_FACTOR) |
135 |
+ predictors[i] * QCELP_LSP_OCTAVE_PREDICTOR |
136 |
+ (i + 1) * ((1 - QCELP_LSP_OCTAVE_PREDICTOR)/11); |
137 |
} |
138 |
smooth = (q->octave_count < 10 ? .875 : 0.1); |
139 |
}else
|
140 |
{ |
141 |
erasure_coeff = QCELP_LSP_OCTAVE_PREDICTOR; |
142 |
|
143 |
assert(q->bitrate == I_F_Q); |
144 |
|
145 |
if(q->erasure_count > 1) |
146 |
erasure_coeff *= (q->erasure_count < 4 ? 0.9 : 0.7); |
147 |
|
148 |
for(i=0; i<10; i++) |
149 |
{ |
150 |
q->predictor_lspf[i] = |
151 |
lspf[i] = (i + 1) * ( 1 - erasure_coeff)/11 |
152 |
+ erasure_coeff * predictors[i]; |
153 |
} |
154 |
smooth = 0.125; |
155 |
} |
156 |
|
157 |
// Check the stability of the LSP frequencies.
|
158 |
lspf[0] = FFMAX(lspf[0], QCELP_LSP_SPREAD_FACTOR); |
159 |
for(i=1; i<10; i++) |
160 |
lspf[i] = FFMAX(lspf[i], (lspf[i-1] + QCELP_LSP_SPREAD_FACTOR));
|
161 |
|
162 |
lspf[9] = FFMIN(lspf[9], (1.0 - QCELP_LSP_SPREAD_FACTOR)); |
163 |
for(i=9; i>0; i--) |
164 |
lspf[i-1] = FFMIN(lspf[i-1], (lspf[i] - QCELP_LSP_SPREAD_FACTOR)); |
165 |
|
166 |
// Low-pass filter the LSP frequencies.
|
167 |
ff_weighted_vector_sumf(lspf, lspf, q->prev_lspf, smooth, 1.0-smooth, 10); |
168 |
}else
|
169 |
{ |
170 |
q->octave_count = 0;
|
171 |
|
172 |
tmp_lspf = 0.;
|
173 |
for(i=0; i<5 ; i++) |
174 |
{ |
175 |
lspf[2*i+0] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][0] * 0.0001; |
176 |
lspf[2*i+1] = tmp_lspf += qcelp_lspvq[i][q->frame.lspv[i]][1] * 0.0001; |
177 |
} |
178 |
|
179 |
// Check for badly received packets.
|
180 |
if(q->bitrate == RATE_QUARTER)
|
181 |
{ |
182 |
if(lspf[9] <= .70 || lspf[9] >= .97) |
183 |
return -1; |
184 |
for(i=3; i<10; i++) |
185 |
if(fabs(lspf[i] - lspf[i-2]) < .08) |
186 |
return -1; |
187 |
}else
|
188 |
{ |
189 |
if(lspf[9] <= .66 || lspf[9] >= .985) |
190 |
return -1; |
191 |
for(i=4; i<10; i++) |
192 |
if (fabs(lspf[i] - lspf[i-4]) < .0931) |
193 |
return -1; |
194 |
} |
195 |
} |
196 |
return 0; |
197 |
} |
198 |
|
199 |
/**
|
200 |
* Convert codebook transmission codes to GAIN and INDEX.
|
201 |
*
|
202 |
* @param q the context
|
203 |
* @param gain array holding the decoded gain
|
204 |
*
|
205 |
* TIA/EIA/IS-733 2.4.6.2
|
206 |
*/
|
207 |
static void decode_gain_and_index(QCELPContext *q, |
208 |
float *gain) {
|
209 |
int i, subframes_count, g1[16]; |
210 |
float slope;
|
211 |
|
212 |
if(q->bitrate >= RATE_QUARTER)
|
213 |
{ |
214 |
switch(q->bitrate)
|
215 |
{ |
216 |
case RATE_FULL: subframes_count = 16; break; |
217 |
case RATE_HALF: subframes_count = 4; break; |
218 |
default: subframes_count = 5; |
219 |
} |
220 |
for(i=0; i<subframes_count; i++) |
221 |
{ |
222 |
g1[i] = 4 * q->frame.cbgain[i];
|
223 |
if(q->bitrate == RATE_FULL && !((i+1) & 3)) |
224 |
{ |
225 |
g1[i] += av_clip((g1[i-1] + g1[i-2] + g1[i-3]) / 3 - 6, 0, 32); |
226 |
} |
227 |
|
228 |
gain[i] = qcelp_g12ga[g1[i]]; |
229 |
|
230 |
if(q->frame.cbsign[i])
|
231 |
{ |
232 |
gain[i] = -gain[i]; |
233 |
q->frame.cindex[i] = (q->frame.cindex[i]-89) & 127; |
234 |
} |
235 |
} |
236 |
|
237 |
q->prev_g1[0] = g1[i-2]; |
238 |
q->prev_g1[1] = g1[i-1]; |
239 |
q->last_codebook_gain = qcelp_g12ga[g1[i-1]];
|
240 |
|
241 |
if(q->bitrate == RATE_QUARTER)
|
242 |
{ |
243 |
// Provide smoothing of the unvoiced excitation energy.
|
244 |
gain[7] = gain[4]; |
245 |
gain[6] = 0.4*gain[3] + 0.6*gain[4]; |
246 |
gain[5] = gain[3]; |
247 |
gain[4] = 0.8*gain[2] + 0.2*gain[3]; |
248 |
gain[3] = 0.2*gain[1] + 0.8*gain[2]; |
249 |
gain[2] = gain[1]; |
250 |
gain[1] = 0.6*gain[0] + 0.4*gain[1]; |
251 |
} |
252 |
}else if (q->bitrate != SILENCE) |
253 |
{ |
254 |
if(q->bitrate == RATE_OCTAVE)
|
255 |
{ |
256 |
g1[0] = 2 * q->frame.cbgain[0] |
257 |
+ av_clip((q->prev_g1[0] + q->prev_g1[1]) / 2 - 5, 0, 54); |
258 |
subframes_count = 8;
|
259 |
}else
|
260 |
{ |
261 |
assert(q->bitrate == I_F_Q); |
262 |
|
263 |
g1[0] = q->prev_g1[1]; |
264 |
switch(q->erasure_count)
|
265 |
{ |
266 |
case 1 : break; |
267 |
case 2 : g1[0] -= 1; break; |
268 |
case 3 : g1[0] -= 2; break; |
269 |
default: g1[0] -= 6; |
270 |
} |
271 |
if(g1[0] < 0) |
272 |
g1[0] = 0; |
273 |
subframes_count = 4;
|
274 |
} |
275 |
// This interpolation is done to produce smoother background noise.
|
276 |
slope = 0.5*(qcelp_g12ga[g1[0]] - q->last_codebook_gain) / subframes_count; |
277 |
for(i=1; i<=subframes_count; i++) |
278 |
gain[i-1] = q->last_codebook_gain + slope * i;
|
279 |
|
280 |
q->last_codebook_gain = gain[i-2];
|
281 |
q->prev_g1[0] = q->prev_g1[1]; |
282 |
q->prev_g1[1] = g1[0]; |
283 |
} |
284 |
} |
285 |
|
286 |
/**
|
287 |
* If the received packet is Rate 1/4 a further sanity check is made of the
|
288 |
* codebook gain.
|
289 |
*
|
290 |
* @param cbgain the unpacked cbgain array
|
291 |
* @return -1 if the sanity check fails, 0 otherwise
|
292 |
*
|
293 |
* TIA/EIA/IS-733 2.4.8.7.3
|
294 |
*/
|
295 |
static int codebook_sanity_check_for_rate_quarter(const uint8_t *cbgain) |
296 |
{ |
297 |
int i, diff, prev_diff=0; |
298 |
|
299 |
for(i=1; i<5; i++) |
300 |
{ |
301 |
diff = cbgain[i] - cbgain[i-1];
|
302 |
if(FFABS(diff) > 10) |
303 |
return -1; |
304 |
else if(FFABS(diff - prev_diff) > 12) |
305 |
return -1; |
306 |
prev_diff = diff; |
307 |
} |
308 |
return 0; |
309 |
} |
310 |
|
311 |
/**
|
312 |
* Compute the scaled codebook vector Cdn From INDEX and GAIN
|
313 |
* for all rates.
|
314 |
*
|
315 |
* The specification lacks some information here.
|
316 |
*
|
317 |
* TIA/EIA/IS-733 has an omission on the codebook index determination
|
318 |
* formula for RATE_FULL and RATE_HALF frames at section 2.4.8.1.1. It says
|
319 |
* you have to subtract the decoded index parameter from the given scaled
|
320 |
* codebook vector index 'n' to get the desired circular codebook index, but
|
321 |
* it does not mention that you have to clamp 'n' to [0-9] in order to get
|
322 |
* RI-compliant results.
|
323 |
*
|
324 |
* The reason for this mistake seems to be the fact they forgot to mention you
|
325 |
* have to do these calculations per codebook subframe and adjust given
|
326 |
* equation values accordingly.
|
327 |
*
|
328 |
* @param q the context
|
329 |
* @param gain array holding the 4 pitch subframe gain values
|
330 |
* @param cdn_vector array for the generated scaled codebook vector
|
331 |
*/
|
332 |
static void compute_svector(QCELPContext *q, const float *gain, |
333 |
float *cdn_vector)
|
334 |
{ |
335 |
int i, j, k;
|
336 |
uint16_t cbseed, cindex; |
337 |
float *rnd, tmp_gain, fir_filter_value;
|
338 |
|
339 |
switch(q->bitrate)
|
340 |
{ |
341 |
case RATE_FULL:
|
342 |
for(i=0; i<16; i++) |
343 |
{ |
344 |
tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; |
345 |
cindex = -q->frame.cindex[i]; |
346 |
for(j=0; j<10; j++) |
347 |
*cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cindex++ & 127];
|
348 |
} |
349 |
break;
|
350 |
case RATE_HALF:
|
351 |
for(i=0; i<4; i++) |
352 |
{ |
353 |
tmp_gain = gain[i] * QCELP_RATE_HALF_CODEBOOK_RATIO; |
354 |
cindex = -q->frame.cindex[i]; |
355 |
for (j = 0; j < 40; j++) |
356 |
*cdn_vector++ = tmp_gain * qcelp_rate_half_codebook[cindex++ & 127];
|
357 |
} |
358 |
break;
|
359 |
case RATE_QUARTER:
|
360 |
cbseed = (0x0003 & q->frame.lspv[4])<<14 | |
361 |
(0x003F & q->frame.lspv[3])<< 8 | |
362 |
(0x0060 & q->frame.lspv[2])<< 1 | |
363 |
(0x0007 & q->frame.lspv[1])<< 3 | |
364 |
(0x0038 & q->frame.lspv[0])>> 3 ; |
365 |
rnd = q->rnd_fir_filter_mem + 20;
|
366 |
for(i=0; i<8; i++) |
367 |
{ |
368 |
tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); |
369 |
for(k=0; k<20; k++) |
370 |
{ |
371 |
cbseed = 521 * cbseed + 259; |
372 |
*rnd = (int16_t)cbseed; |
373 |
|
374 |
// FIR filter
|
375 |
fir_filter_value = 0.0; |
376 |
for(j=0; j<10; j++) |
377 |
fir_filter_value += qcelp_rnd_fir_coefs[j ] |
378 |
* (rnd[-j ] + rnd[-20+j]);
|
379 |
|
380 |
fir_filter_value += qcelp_rnd_fir_coefs[10] * rnd[-10]; |
381 |
*cdn_vector++ = tmp_gain * fir_filter_value; |
382 |
rnd++; |
383 |
} |
384 |
} |
385 |
memcpy(q->rnd_fir_filter_mem, q->rnd_fir_filter_mem + 160, 20 * sizeof(float)); |
386 |
break;
|
387 |
case RATE_OCTAVE:
|
388 |
cbseed = q->first16bits; |
389 |
for(i=0; i<8; i++) |
390 |
{ |
391 |
tmp_gain = gain[i] * (QCELP_SQRT1887 / 32768.0); |
392 |
for(j=0; j<20; j++) |
393 |
{ |
394 |
cbseed = 521 * cbseed + 259; |
395 |
*cdn_vector++ = tmp_gain * (int16_t)cbseed; |
396 |
} |
397 |
} |
398 |
break;
|
399 |
case I_F_Q:
|
400 |
cbseed = -44; // random codebook index |
401 |
for(i=0; i<4; i++) |
402 |
{ |
403 |
tmp_gain = gain[i] * QCELP_RATE_FULL_CODEBOOK_RATIO; |
404 |
for(j=0; j<40; j++) |
405 |
*cdn_vector++ = tmp_gain * qcelp_rate_full_codebook[cbseed++ & 127];
|
406 |
} |
407 |
break;
|
408 |
case SILENCE:
|
409 |
memset(cdn_vector, 0, 160 * sizeof(float)); |
410 |
break;
|
411 |
} |
412 |
} |
413 |
|
414 |
/**
|
415 |
* Apply generic gain control.
|
416 |
*
|
417 |
* @param v_out output vector
|
418 |
* @param v_in gain-controlled vector
|
419 |
* @param v_ref vector to control gain of
|
420 |
*
|
421 |
* TIA/EIA/IS-733 2.4.8.3, 2.4.8.6
|
422 |
*/
|
423 |
static void apply_gain_ctrl(float *v_out, const float *v_ref, |
424 |
const float *v_in) |
425 |
{ |
426 |
int i;
|
427 |
|
428 |
for (i = 0; i < 160; i += 40) |
429 |
ff_scale_vector_to_given_sum_of_squares(v_out + i, v_in + i, |
430 |
ff_dot_productf(v_ref + i, |
431 |
v_ref + i, 40),
|
432 |
40);
|
433 |
} |
434 |
|
435 |
/**
|
436 |
* Apply filter in pitch-subframe steps.
|
437 |
*
|
438 |
* @param memory buffer for the previous state of the filter
|
439 |
* - must be able to contain 303 elements
|
440 |
* - the 143 first elements are from the previous state
|
441 |
* - the next 160 are for output
|
442 |
* @param v_in input filter vector
|
443 |
* @param gain per-subframe gain array, each element is between 0.0 and 2.0
|
444 |
* @param lag per-subframe lag array, each element is
|
445 |
* - between 16 and 143 if its corresponding pfrac is 0,
|
446 |
* - between 16 and 139 otherwise
|
447 |
* @param pfrac per-subframe boolean array, 1 if the lag is fractional, 0
|
448 |
* otherwise
|
449 |
*
|
450 |
* @return filter output vector
|
451 |
*/
|
452 |
static const float *do_pitchfilter(float memory[303], const float v_in[160], |
453 |
const float gain[4], const uint8_t *lag, |
454 |
const uint8_t pfrac[4]) |
455 |
{ |
456 |
int i, j;
|
457 |
float *v_lag, *v_out;
|
458 |
const float *v_len; |
459 |
|
460 |
v_out = memory + 143; // Output vector starts at memory[143]. |
461 |
|
462 |
for(i=0; i<4; i++) |
463 |
{ |
464 |
if(gain[i])
|
465 |
{ |
466 |
v_lag = memory + 143 + 40 * i - lag[i]; |
467 |
for(v_len=v_in+40; v_in<v_len; v_in++) |
468 |
{ |
469 |
if(pfrac[i]) // If it is a fractional lag... |
470 |
{ |
471 |
for(j=0, *v_out=0.; j<4; j++) |
472 |
*v_out += qcelp_hammsinc_table[j] * (v_lag[j-4] + v_lag[3-j]); |
473 |
}else
|
474 |
*v_out = *v_lag; |
475 |
|
476 |
*v_out = *v_in + gain[i] * *v_out; |
477 |
|
478 |
v_lag++; |
479 |
v_out++; |
480 |
} |
481 |
}else
|
482 |
{ |
483 |
memcpy(v_out, v_in, 40 * sizeof(float)); |
484 |
v_in += 40;
|
485 |
v_out += 40;
|
486 |
} |
487 |
} |
488 |
|
489 |
memmove(memory, memory + 160, 143 * sizeof(float)); |
490 |
return memory + 143; |
491 |
} |
492 |
|
493 |
/**
|
494 |
* Apply pitch synthesis filter and pitch prefilter to the scaled codebook vector.
|
495 |
* TIA/EIA/IS-733 2.4.5.2, 2.4.8.7.2
|
496 |
*
|
497 |
* @param q the context
|
498 |
* @param cdn_vector the scaled codebook vector
|
499 |
*/
|
500 |
static void apply_pitch_filters(QCELPContext *q, float *cdn_vector) |
501 |
{ |
502 |
int i;
|
503 |
const float *v_synthesis_filtered, *v_pre_filtered; |
504 |
|
505 |
if(q->bitrate >= RATE_HALF ||
|
506 |
q->bitrate == SILENCE || |
507 |
(q->bitrate == I_F_Q && (q->prev_bitrate >= RATE_HALF))) |
508 |
{ |
509 |
|
510 |
if(q->bitrate >= RATE_HALF)
|
511 |
{ |
512 |
|
513 |
// Compute gain & lag for the whole frame.
|
514 |
for(i=0; i<4; i++) |
515 |
{ |
516 |
q->pitch_gain[i] = q->frame.plag[i] ? (q->frame.pgain[i] + 1) * 0.25 : 0.0; |
517 |
|
518 |
q->pitch_lag[i] = q->frame.plag[i] + 16;
|
519 |
} |
520 |
}else
|
521 |
{ |
522 |
float max_pitch_gain;
|
523 |
|
524 |
if (q->bitrate == I_F_Q)
|
525 |
{ |
526 |
if (q->erasure_count < 3) |
527 |
max_pitch_gain = 0.9 - 0.3 * (q->erasure_count - 1); |
528 |
else
|
529 |
max_pitch_gain = 0.0; |
530 |
}else
|
531 |
{ |
532 |
assert(q->bitrate == SILENCE); |
533 |
max_pitch_gain = 1.0; |
534 |
} |
535 |
for(i=0; i<4; i++) |
536 |
q->pitch_gain[i] = FFMIN(q->pitch_gain[i], max_pitch_gain); |
537 |
|
538 |
memset(q->frame.pfrac, 0, sizeof(q->frame.pfrac)); |
539 |
} |
540 |
|
541 |
// pitch synthesis filter
|
542 |
v_synthesis_filtered = do_pitchfilter(q->pitch_synthesis_filter_mem, |
543 |
cdn_vector, q->pitch_gain, |
544 |
q->pitch_lag, q->frame.pfrac); |
545 |
|
546 |
// pitch prefilter update
|
547 |
for(i=0; i<4; i++) |
548 |
q->pitch_gain[i] = 0.5 * FFMIN(q->pitch_gain[i], 1.0); |
549 |
|
550 |
v_pre_filtered = do_pitchfilter(q->pitch_pre_filter_mem, |
551 |
v_synthesis_filtered, |
552 |
q->pitch_gain, q->pitch_lag, |
553 |
q->frame.pfrac); |
554 |
|
555 |
apply_gain_ctrl(cdn_vector, v_synthesis_filtered, v_pre_filtered); |
556 |
}else
|
557 |
{ |
558 |
memcpy(q->pitch_synthesis_filter_mem, cdn_vector + 17,
|
559 |
143 * sizeof(float)); |
560 |
memcpy(q->pitch_pre_filter_mem, cdn_vector + 17, 143 * sizeof(float)); |
561 |
memset(q->pitch_gain, 0, sizeof(q->pitch_gain)); |
562 |
memset(q->pitch_lag, 0, sizeof(q->pitch_lag)); |
563 |
} |
564 |
} |
565 |
|
566 |
/**
|
567 |
* Reconstruct LPC coefficients from the line spectral pair frequencies
|
568 |
* and perform bandwidth expansion.
|
569 |
*
|
570 |
* @param lspf line spectral pair frequencies
|
571 |
* @param lpc linear predictive coding coefficients
|
572 |
*
|
573 |
* @note: bandwidth_expansion_coeff could be precalculated into a table
|
574 |
* but it seems to be slower on x86
|
575 |
*
|
576 |
* TIA/EIA/IS-733 2.4.3.3.5
|
577 |
*/
|
578 |
static void lspf2lpc(const float *lspf, float *lpc) |
579 |
{ |
580 |
double lsp[10]; |
581 |
double bandwidth_expansion_coeff = QCELP_BANDWIDTH_EXPANSION_COEFF;
|
582 |
int i;
|
583 |
|
584 |
for (i=0; i<10; i++) |
585 |
lsp[i] = cos(M_PI * lspf[i]); |
586 |
|
587 |
ff_acelp_lspd2lpc(lsp, lpc, 5);
|
588 |
|
589 |
for (i=0; i<10; i++) |
590 |
{ |
591 |
lpc[i] *= bandwidth_expansion_coeff; |
592 |
bandwidth_expansion_coeff *= QCELP_BANDWIDTH_EXPANSION_COEFF; |
593 |
} |
594 |
} |
595 |
|
596 |
/**
|
597 |
* Interpolate LSP frequencies and compute LPC coefficients
|
598 |
* for a given bitrate & pitch subframe.
|
599 |
*
|
600 |
* TIA/EIA/IS-733 2.4.3.3.4, 2.4.8.7.2
|
601 |
*
|
602 |
* @param q the context
|
603 |
* @param curr_lspf LSP frequencies vector of the current frame
|
604 |
* @param lpc float vector for the resulting LPC
|
605 |
* @param subframe_num frame number in decoded stream
|
606 |
*/
|
607 |
static void interpolate_lpc(QCELPContext *q, const float *curr_lspf, |
608 |
float *lpc, const int subframe_num) |
609 |
{ |
610 |
float interpolated_lspf[10]; |
611 |
float weight;
|
612 |
|
613 |
if(q->bitrate >= RATE_QUARTER)
|
614 |
weight = 0.25 * (subframe_num + 1); |
615 |
else if(q->bitrate == RATE_OCTAVE && !subframe_num) |
616 |
weight = 0.625; |
617 |
else
|
618 |
weight = 1.0; |
619 |
|
620 |
if(weight != 1.0) |
621 |
{ |
622 |
ff_weighted_vector_sumf(interpolated_lspf, curr_lspf, q->prev_lspf, |
623 |
weight, 1.0 - weight, 10); |
624 |
lspf2lpc(interpolated_lspf, lpc); |
625 |
}else if(q->bitrate >= RATE_QUARTER || |
626 |
(q->bitrate == I_F_Q && !subframe_num)) |
627 |
lspf2lpc(curr_lspf, lpc); |
628 |
else if(q->bitrate == SILENCE && !subframe_num) |
629 |
lspf2lpc(q->prev_lspf, lpc); |
630 |
} |
631 |
|
632 |
static qcelp_packet_rate buf_size2bitrate(const int buf_size) |
633 |
{ |
634 |
switch(buf_size)
|
635 |
{ |
636 |
case 35: return RATE_FULL; |
637 |
case 17: return RATE_HALF; |
638 |
case 8: return RATE_QUARTER; |
639 |
case 4: return RATE_OCTAVE; |
640 |
case 1: return SILENCE; |
641 |
} |
642 |
|
643 |
return I_F_Q;
|
644 |
} |
645 |
|
646 |
/**
|
647 |
* Determine the bitrate from the frame size and/or the first byte of the frame.
|
648 |
*
|
649 |
* @param avctx the AV codec context
|
650 |
* @param buf_size length of the buffer
|
651 |
* @param buf the bufffer
|
652 |
*
|
653 |
* @return the bitrate on success,
|
654 |
* I_F_Q if the bitrate cannot be satisfactorily determined
|
655 |
*
|
656 |
* TIA/EIA/IS-733 2.4.8.7.1
|
657 |
*/
|
658 |
static qcelp_packet_rate determine_bitrate(AVCodecContext *avctx, const int buf_size, |
659 |
const uint8_t **buf)
|
660 |
{ |
661 |
qcelp_packet_rate bitrate; |
662 |
|
663 |
if((bitrate = buf_size2bitrate(buf_size)) >= 0) |
664 |
{ |
665 |
if(bitrate > **buf)
|
666 |
{ |
667 |
QCELPContext *q = avctx->priv_data; |
668 |
if (!q->warned_buf_mismatch_bitrate)
|
669 |
{ |
670 |
av_log(avctx, AV_LOG_WARNING, |
671 |
"Claimed bitrate and buffer size mismatch.\n");
|
672 |
q->warned_buf_mismatch_bitrate = 1;
|
673 |
} |
674 |
bitrate = **buf; |
675 |
}else if(bitrate < **buf) |
676 |
{ |
677 |
av_log(avctx, AV_LOG_ERROR, |
678 |
"Buffer is too small for the claimed bitrate.\n");
|
679 |
return I_F_Q;
|
680 |
} |
681 |
(*buf)++; |
682 |
}else if((bitrate = buf_size2bitrate(buf_size + 1)) >= 0) |
683 |
{ |
684 |
av_log(avctx, AV_LOG_WARNING, |
685 |
"Bitrate byte is missing, guessing the bitrate from packet size.\n");
|
686 |
}else
|
687 |
return I_F_Q;
|
688 |
|
689 |
if(bitrate == SILENCE)
|
690 |
{ |
691 |
//FIXME: Remove experimental warning when tested with samples.
|
692 |
av_log_ask_for_sample(avctx, "'Blank frame handling is experimental.");
|
693 |
} |
694 |
return bitrate;
|
695 |
} |
696 |
|
697 |
static void warn_insufficient_frame_quality(AVCodecContext *avctx, |
698 |
const char *message) |
699 |
{ |
700 |
av_log(avctx, AV_LOG_WARNING, "Frame #%d, IFQ: %s\n", avctx->frame_number,
|
701 |
message); |
702 |
} |
703 |
|
704 |
static void postfilter(QCELPContext *q, float *samples, float *lpc) |
705 |
{ |
706 |
static const float pow_0_775[10] = { |
707 |
0.775000, 0.600625, 0.465484, 0.360750, 0.279582, |
708 |
0.216676, 0.167924, 0.130141, 0.100859, 0.078166 |
709 |
}, pow_0_625[10] = {
|
710 |
0.625000, 0.390625, 0.244141, 0.152588, 0.095367, |
711 |
0.059605, 0.037253, 0.023283, 0.014552, 0.009095 |
712 |
}; |
713 |
float lpc_s[10], lpc_p[10], pole_out[170], zero_out[160]; |
714 |
int n;
|
715 |
|
716 |
for (n = 0; n < 10; n++) { |
717 |
lpc_s[n] = lpc[n] * pow_0_625[n]; |
718 |
lpc_p[n] = lpc[n] * pow_0_775[n]; |
719 |
} |
720 |
|
721 |
ff_celp_lp_zero_synthesis_filterf(zero_out, lpc_s, |
722 |
q->formant_mem + 10, 160, 10); |
723 |
memcpy(pole_out, q->postfilter_synth_mem, sizeof(float) * 10); |
724 |
ff_celp_lp_synthesis_filterf(pole_out + 10, lpc_p, zero_out, 160, 10); |
725 |
memcpy(q->postfilter_synth_mem, pole_out + 160, sizeof(float) * 10); |
726 |
|
727 |
ff_tilt_compensation(&q->postfilter_tilt_mem, 0.3, pole_out + 10, 160); |
728 |
|
729 |
ff_adaptive_gain_control(samples, pole_out + 10,
|
730 |
ff_dot_productf(q->formant_mem + 10, q->formant_mem + 10, 160), |
731 |
160, 0.9375, &q->postfilter_agc_mem); |
732 |
} |
733 |
|
734 |
static int qcelp_decode_frame(AVCodecContext *avctx, void *data, int *data_size, |
735 |
AVPacket *avpkt) |
736 |
{ |
737 |
const uint8_t *buf = avpkt->data;
|
738 |
int buf_size = avpkt->size;
|
739 |
QCELPContext *q = avctx->priv_data; |
740 |
float *outbuffer = data;
|
741 |
int i;
|
742 |
float quantized_lspf[10], lpc[10]; |
743 |
float gain[16]; |
744 |
float *formant_mem;
|
745 |
|
746 |
if((q->bitrate = determine_bitrate(avctx, buf_size, &buf)) == I_F_Q)
|
747 |
{ |
748 |
warn_insufficient_frame_quality(avctx, "bitrate cannot be determined.");
|
749 |
goto erasure;
|
750 |
} |
751 |
|
752 |
if(q->bitrate == RATE_OCTAVE &&
|
753 |
(q->first16bits = AV_RB16(buf)) == 0xFFFF)
|
754 |
{ |
755 |
warn_insufficient_frame_quality(avctx, "Bitrate is 1/8 and first 16 bits are on.");
|
756 |
goto erasure;
|
757 |
} |
758 |
|
759 |
if(q->bitrate > SILENCE)
|
760 |
{ |
761 |
const QCELPBitmap *bitmaps = qcelp_unpacking_bitmaps_per_rate[q->bitrate];
|
762 |
const QCELPBitmap *bitmaps_end = qcelp_unpacking_bitmaps_per_rate[q->bitrate]
|
763 |
+ qcelp_unpacking_bitmaps_lengths[q->bitrate]; |
764 |
uint8_t *unpacked_data = (uint8_t *)&q->frame; |
765 |
|
766 |
init_get_bits(&q->gb, buf, 8*buf_size);
|
767 |
|
768 |
memset(&q->frame, 0, sizeof(QCELPFrame)); |
769 |
|
770 |
for(; bitmaps < bitmaps_end; bitmaps++)
|
771 |
unpacked_data[bitmaps->index] |= get_bits(&q->gb, bitmaps->bitlen) << bitmaps->bitpos; |
772 |
|
773 |
// Check for erasures/blanks on rates 1, 1/4 and 1/8.
|
774 |
if(q->frame.reserved)
|
775 |
{ |
776 |
warn_insufficient_frame_quality(avctx, "Wrong data in reserved frame area.");
|
777 |
goto erasure;
|
778 |
} |
779 |
if(q->bitrate == RATE_QUARTER &&
|
780 |
codebook_sanity_check_for_rate_quarter(q->frame.cbgain)) |
781 |
{ |
782 |
warn_insufficient_frame_quality(avctx, "Codebook gain sanity check failed.");
|
783 |
goto erasure;
|
784 |
} |
785 |
|
786 |
if(q->bitrate >= RATE_HALF)
|
787 |
{ |
788 |
for(i=0; i<4; i++) |
789 |
{ |
790 |
if(q->frame.pfrac[i] && q->frame.plag[i] >= 124) |
791 |
{ |
792 |
warn_insufficient_frame_quality(avctx, "Cannot initialize pitch filter.");
|
793 |
goto erasure;
|
794 |
} |
795 |
} |
796 |
} |
797 |
} |
798 |
|
799 |
decode_gain_and_index(q, gain); |
800 |
compute_svector(q, gain, outbuffer); |
801 |
|
802 |
if(decode_lspf(q, quantized_lspf) < 0) |
803 |
{ |
804 |
warn_insufficient_frame_quality(avctx, "Badly received packets in frame.");
|
805 |
goto erasure;
|
806 |
} |
807 |
|
808 |
|
809 |
apply_pitch_filters(q, outbuffer); |
810 |
|
811 |
if(q->bitrate == I_F_Q)
|
812 |
{ |
813 |
erasure:
|
814 |
q->bitrate = I_F_Q; |
815 |
q->erasure_count++; |
816 |
decode_gain_and_index(q, gain); |
817 |
compute_svector(q, gain, outbuffer); |
818 |
decode_lspf(q, quantized_lspf); |
819 |
apply_pitch_filters(q, outbuffer); |
820 |
}else
|
821 |
q->erasure_count = 0;
|
822 |
|
823 |
formant_mem = q->formant_mem + 10;
|
824 |
for(i=0; i<4; i++) |
825 |
{ |
826 |
interpolate_lpc(q, quantized_lspf, lpc, i); |
827 |
ff_celp_lp_synthesis_filterf(formant_mem, lpc, outbuffer + i * 40, 40, |
828 |
10);
|
829 |
formant_mem += 40;
|
830 |
} |
831 |
|
832 |
// postfilter, as per TIA/EIA/IS-733 2.4.8.6
|
833 |
postfilter(q, outbuffer, lpc); |
834 |
|
835 |
memcpy(q->formant_mem, q->formant_mem + 160, 10 * sizeof(float)); |
836 |
|
837 |
memcpy(q->prev_lspf, quantized_lspf, sizeof(q->prev_lspf));
|
838 |
q->prev_bitrate = q->bitrate; |
839 |
|
840 |
*data_size = 160 * sizeof(*outbuffer); |
841 |
|
842 |
return *data_size;
|
843 |
} |
844 |
|
845 |
AVCodec ff_qcelp_decoder = |
846 |
{ |
847 |
.name = "qcelp",
|
848 |
.type = AVMEDIA_TYPE_AUDIO, |
849 |
.id = CODEC_ID_QCELP, |
850 |
.init = qcelp_decode_init, |
851 |
.decode = qcelp_decode_frame, |
852 |
.priv_data_size = sizeof(QCELPContext),
|
853 |
.long_name = NULL_IF_CONFIG_SMALL("QCELP / PureVoice"),
|
854 |
}; |