| 1 |
/* |
| 2 |
* TwoLAME: an optimized MPEG Audio Layer Two encoder |
| 3 |
* |
| 4 |
* Copyright (C) 2001-2004 Michael Cheng |
| 5 |
* Copyright (C) 2004-2018 The TwoLAME Project |
| 6 |
* |
| 7 |
* This library 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 |
* This library 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 this library; if not, write to the Free Software |
| 19 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 20 |
* |
| 21 |
*/ |
| 22 |
|
| 23 |
|
| 24 |
#include <stdio.h> |
| 25 |
#include <stdlib.h> |
| 26 |
#include <math.h> |
| 27 |
|
| 28 |
#include "twolame.h" |
| 29 |
#include "common.h" |
| 30 |
#include "mem.h" |
| 31 |
#include "fft.h" |
| 32 |
#include "psycho_1.h" |
| 33 |
|
| 34 |
/********************************************************************** |
| 35 |
|
| 36 |
This module implements the psychoacoustic model I for the |
| 37 |
MPEG encoder layer II. It uses simplified tonal and noise masking |
| 38 |
threshold analysis to generate SMR for the encoder bit allocation |
| 39 |
routine. |
| 40 |
|
| 41 |
**********************************************************************/ |
| 42 |
|
| 43 |
|
| 44 |
static int *psycho_1_read_cbound(int lay, int freq, int *crit_band) |
| 45 |
/* this function reads in critical band boundaries */ |
| 46 |
{ |
| 47 |
|
| 48 |
#include "psycho_1_critband.h" |
| 49 |
int *cbound; |
| 50 |
int i, k; |
| 51 |
|
| 52 |
if ((lay < 1) || (lay > 2)) { |
| 53 |
fprintf(stderr, "Internal error (read_cbound())\n"); |
| 54 |
return (NULL); |
| 55 |
} |
| 56 |
if ((freq < 0) || (freq > 6) || (freq == 3)) { |
| 57 |
fprintf(stderr, "Internal error (read_cbound())\n"); |
| 58 |
return (NULL); |
| 59 |
} |
| 60 |
|
| 61 |
*crit_band = SecondCriticalBand[freq][0]; |
| 62 |
cbound = (int *) TWOLAME_MALLOC(sizeof(int) * *crit_band); |
| 63 |
for (i = 0; i < *crit_band; i++) { |
| 64 |
k = SecondCriticalBand[freq][i + 1]; |
| 65 |
if (k != 0) { |
| 66 |
cbound[i] = k; |
| 67 |
} else { |
| 68 |
fprintf(stderr, "Internal error (read_cbound())\n"); |
| 69 |
return (NULL); |
| 70 |
} |
| 71 |
} |
| 72 |
return (cbound); |
| 73 |
} |
| 74 |
|
| 75 |
/* reads in the frequency bands and bark values */ |
| 76 |
static void psycho_1_read_freq_band(g_ptr * ltg, int lay, int freq, int *sub_size) |
| 77 |
{ |
| 78 |
|
| 79 |
#include "psycho_1_freqtable.h" |
| 80 |
|
| 81 |
int i, k; |
| 82 |
|
| 83 |
if ((freq < 0) || (freq > 6) || (freq == 3)) { |
| 84 |
fprintf(stderr, "Internal error (read_freq_band())\n"); |
| 85 |
return; |
| 86 |
} |
| 87 |
|
| 88 |
/* read input for freq. subbands */ |
| 89 |
|
| 90 |
*sub_size = SecondFreqEntries[freq] + 1; |
| 91 |
*ltg = (g_ptr) TWOLAME_MALLOC(sizeof(g_thres) * *sub_size); |
| 92 |
(*ltg)[0].line = 0; /* initialize global masking threshold */ |
| 93 |
(*ltg)[0].bark = 0.0; |
| 94 |
(*ltg)[0].hear = 0.0; |
| 95 |
for (i = 1; i < *sub_size; i++) { |
| 96 |
k = SecondFreqSubband[freq][i - 1].line; |
| 97 |
if (k != 0) { |
| 98 |
(*ltg)[i].line = k; |
| 99 |
(*ltg)[i].bark = SecondFreqSubband[freq][i - 1].bark; |
| 100 |
(*ltg)[i].hear = SecondFreqSubband[freq][i - 1].hear; |
| 101 |
} else { |
| 102 |
fprintf(stderr, "Internal error (read_freq_band())\n"); |
| 103 |
return; |
| 104 |
} |
| 105 |
} |
| 106 |
} |
| 107 |
|
| 108 |
|
| 109 |
static void psycho_1_make_map(int sub_size, mask power[HAN_SIZE], g_thres * ltg) |
| 110 |
/* this function calculates the global masking threshold */ |
| 111 |
{ |
| 112 |
int i, j; |
| 113 |
|
| 114 |
for (i = 1; i < sub_size; i++) |
| 115 |
for (j = ltg[i - 1].line; j <= ltg[i].line; j++) |
| 116 |
power[j].map = i; |
| 117 |
} |
| 118 |
|
| 119 |
static void psycho_1_init_add_db(psycho_1_mem * mem) |
| 120 |
{ |
| 121 |
int i; |
| 122 |
FLOAT x; |
| 123 |
for (i = 0; i < DBTAB; i++) { |
| 124 |
x = (FLOAT) i / 10.0; |
| 125 |
mem->dbtable[i] = 10 * log10(1 + pow(10.0, x / 10.0)) - x; |
| 126 |
} |
| 127 |
} |
| 128 |
|
| 129 |
static inline FLOAT add_db(psycho_1_mem * mem, FLOAT a, FLOAT b) |
| 130 |
{ |
| 131 |
/* MFC - if the difference between a and b is large (>99), then just return the largest one. |
| 132 |
(about 10% of the time) - For differences between 0 and 99, return the largest value, but |
| 133 |
add in a pre-calculated difference value. - the value 99 was chosen arbitarily. - maximum |
| 134 |
(a-b) i've seen is 572 */ |
| 135 |
FLOAT fdiff; |
| 136 |
int idiff; |
| 137 |
fdiff = (10.0 * (a - b)); |
| 138 |
|
| 139 |
if (fdiff > 990.0) { |
| 140 |
return a; |
| 141 |
} |
| 142 |
if (fdiff < -990.0) { |
| 143 |
return (b); |
| 144 |
} |
| 145 |
|
| 146 |
idiff = (int) fdiff; |
| 147 |
if (idiff >= 0) { |
| 148 |
return (a + mem->dbtable[idiff]); |
| 149 |
} |
| 150 |
|
| 151 |
return (b + mem->dbtable[-idiff]); |
| 152 |
} |
| 153 |
|
| 154 |
/**************************************************************** |
| 155 |
* Window the samples then, |
| 156 |
* Fast Fourier transform of the input samples. |
| 157 |
* |
| 158 |
* ( call the FHT-based fft() in fft.c ) |
| 159 |
* |
| 160 |
* |
| 161 |
****************************************************************/ |
| 162 |
static void psycho_1_hann_fft_pickmax(FLOAT sample[FFT_SIZE], mask power[HAN_SIZE], |
| 163 |
FLOAT spike[SBLIMIT], FLOAT energy[FFT_SIZE]) |
| 164 |
{ |
| 165 |
FLOAT x_real[FFT_SIZE]; |
| 166 |
register int i, j; |
| 167 |
register FLOAT sqrt_8_over_3; |
| 168 |
static int init = 0; |
| 169 |
static FLOAT window[FFT_SIZE]; |
| 170 |
FLOAT sum; |
| 171 |
|
| 172 |
if (!init) { |
| 173 |
/* calculate window function for the Fourier transform */ |
| 174 |
/* These values need only be initiliased once, regardless of the caller */ |
| 175 |
sqrt_8_over_3 = pow(8.0 / 3.0, 0.5); |
| 176 |
for (i = 0; i < FFT_SIZE; i++) { |
| 177 |
/* Hann window formula */ |
| 178 |
window[i] = sqrt_8_over_3 * 0.5 * (1 - cos(2.0 * PI * i / (FFT_SIZE))) / FFT_SIZE; |
| 179 |
} |
| 180 |
init = 1; |
| 181 |
} |
| 182 |
for (i = 0; i < FFT_SIZE; i++) |
| 183 |
x_real[i] = (FLOAT) (sample[i] * window[i]); |
| 184 |
|
| 185 |
twolame_psycho_1_fft(x_real, energy, FFT_SIZE); |
| 186 |
|
| 187 |
for (i = 0; i < HAN_SIZE; i++) { /* calculate power density spectrum */ |
| 188 |
if (energy[i] < 1E-20) |
| 189 |
power[i].x = -200.0 + POWERNORM; |
| 190 |
else |
| 191 |
power[i].x = 10 * log10(energy[i]) + POWERNORM; |
| 192 |
power[i].next = STOP; |
| 193 |
power[i].type = FALSE; |
| 194 |
} |
| 195 |
|
| 196 |
/* Calculate the sum of spectral component in each subband from bound 4-16 */ |
| 197 |
|
| 198 |
#define CF 1073741824 /* pow(10, 0.1*POWERNORM) */ |
| 199 |
#define DBM 1E-20 /* pow(10.0, 0.1*DBMIN */ |
| 200 |
for (i = 0; i < HAN_SIZE; spike[i >> 4] = 10.0 * log10(sum), i += 16) { |
| 201 |
for (j = 0, sum = DBM; j < 16; j++) |
| 202 |
sum += CF * energy[i + j]; |
| 203 |
} |
| 204 |
} |
| 205 |
|
| 206 |
/**************************************************************** |
| 207 |
* |
| 208 |
* This function labels the tonal component in the power |
| 209 |
* spectrum. |
| 210 |
* |
| 211 |
****************************************************************/ |
| 212 |
|
| 213 |
static void psycho_1_tonal_label(psycho_1_mem * mem, int *tone) |
| 214 |
/* this function extracts (tonal) sinusoidals from the spectrum */ |
| 215 |
{ |
| 216 |
int i, j, last = LAST, first, run, last_but_one = LAST; /* dpwe */ |
| 217 |
FLOAT max; |
| 218 |
mask *power = mem->power; |
| 219 |
|
| 220 |
*tone = LAST; |
| 221 |
for (i = 2; i < HAN_SIZE - 12; i++) { |
| 222 |
if (power[i].x > power[i - 1].x && power[i].x >= power[i + 1].x) { |
| 223 |
power[i].type = TONE; |
| 224 |
power[i].next = LAST; |
| 225 |
if (last != LAST) |
| 226 |
power[last].next = i; |
| 227 |
else |
| 228 |
first = *tone = i; |
| 229 |
last = i; |
| 230 |
} |
| 231 |
} |
| 232 |
last = LAST; |
| 233 |
first = *tone; |
| 234 |
*tone = LAST; |
| 235 |
while ((first != LAST) && (first != STOP)) { /* the conditions for the tonal */ |
| 236 |
if (first < 3 || first > 500) |
| 237 |
run = 0; /* otherwise k+/-j will be out of bounds */ |
| 238 |
else if (first < 63) |
| 239 |
run = 2; /* components in layer II, which */ |
| 240 |
else if (first < 127) |
| 241 |
run = 3; /* are the boundaries for calc. */ |
| 242 |
else if (first < 255) |
| 243 |
run = 6; /* the tonal components */ |
| 244 |
else |
| 245 |
run = 12; |
| 246 |
max = power[first].x - 7; /* after calculation of tonal */ |
| 247 |
for (j = 2; j <= run; j++) /* components, set to local max */ |
| 248 |
if (max < power[first - j].x || max < power[first + j].x) { |
| 249 |
power[first].type = FALSE; |
| 250 |
break; |
| 251 |
} |
| 252 |
if (power[first].type == TONE) { /* extract tonal components */ |
| 253 |
int help = first; |
| 254 |
if (*tone == LAST) |
| 255 |
*tone = first; |
| 256 |
while ((power[help].next != LAST) && (power[help].next - first) <= run) |
| 257 |
help = power[help].next; |
| 258 |
help = power[help].next; |
| 259 |
power[first].next = help; |
| 260 |
if ((first - last) <= run) { |
| 261 |
if (last_but_one != LAST) |
| 262 |
power[last_but_one].next = first; |
| 263 |
} |
| 264 |
if (first > 1 && first < 500) { /* calculate the sum of the */ |
| 265 |
FLOAT tmp; /* powers of the components */ |
| 266 |
tmp = add_db(mem, power[first - 1].x, power[first + 1].x); |
| 267 |
power[first].x = add_db(mem, power[first].x, tmp); |
| 268 |
} |
| 269 |
for (j = 1; j <= run; j++) { |
| 270 |
power[first - j].x = power[first + j].x = DBMIN; |
| 271 |
power[first - j].next = power[first + j].next = STOP; |
| 272 |
power[first - j].type = power[first + j].type = FALSE; |
| 273 |
} |
| 274 |
last_but_one = last; |
| 275 |
last = first; |
| 276 |
first = power[first].next; |
| 277 |
} else { |
| 278 |
int ll; |
| 279 |
if (last == LAST); /* *tone = power[first].next; dpwe */ |
| 280 |
else |
| 281 |
power[last].next = power[first].next; |
| 282 |
ll = first; |
| 283 |
first = power[first].next; |
| 284 |
power[ll].next = STOP; |
| 285 |
} |
| 286 |
} |
| 287 |
} |
| 288 |
|
| 289 |
/**************************************************************** |
| 290 |
* |
| 291 |
* This function groups all the remaining non-tonal |
| 292 |
* spectral lines into critical band where they are replaced by |
| 293 |
* one single line. |
| 294 |
* |
| 295 |
****************************************************************/ |
| 296 |
|
| 297 |
static void psycho_1_noise_label(psycho_1_mem * mem, int *noise, FLOAT energy[FFT_SIZE]) |
| 298 |
{ |
| 299 |
int i, j, centre, last = LAST; |
| 300 |
FLOAT index, weight, sum; |
| 301 |
int crit_band = mem->crit_band; |
| 302 |
int *cbound = mem->cbound; |
| 303 |
mask *power = mem->power; |
| 304 |
/* calculate the remaining spectral */ |
| 305 |
for (i = 0; i < crit_band - 1; i++) { /* lines for non-tonal components */ |
| 306 |
for (j = cbound[i], weight = 0.0, sum = DBMIN; j < cbound[i + 1]; j++) { |
| 307 |
if (power[j].type != TONE) { |
| 308 |
if (power[j].x != DBMIN) { |
| 309 |
sum = add_db(mem, power[j].x, sum); |
| 310 |
/* Weight is used in finding the geometric mean of the noise energy within a |
| 311 |
subband */ |
| 312 |
weight += CF * energy[j] * (FLOAT) (j - cbound[i]) / (FLOAT) (cbound[i + 1] - cbound[i]); /* correction |
| 313 |
*/ |
| 314 |
power[j].x = DBMIN; |
| 315 |
} |
| 316 |
} /* check to see if the spectral line is low dB, and if */ |
| 317 |
} /* so replace the center of the critical band, which is */ |
| 318 |
/* the center freq. of the noise component */ |
| 319 |
|
| 320 |
if (sum <= DBMIN) |
| 321 |
centre = (cbound[i + 1] + cbound[i]) / 2; |
| 322 |
else { |
| 323 |
/* fprintf(stderr, "%i [%f %f] -", count++,weight/pow(10.0,0.1*sum), |
| 324 |
weight*pow(10.0,-0.1*sum)); */ |
| 325 |
index = weight * pow(10.0, -0.1 * sum); |
| 326 |
centre = cbound[i] + (int) (index * (FLOAT) (cbound[i + 1] - cbound[i])); |
| 327 |
} |
| 328 |
|
| 329 |
|
| 330 |
/* locate next non-tonal component until finished; */ |
| 331 |
/* add to list of non-tonal components */ |
| 332 |
|
| 333 |
/* Masahiro Iwadare's fix for infinite looping problem? */ |
| 334 |
if (power[centre].type == TONE) { |
| 335 |
if (power[centre + 1].type == TONE) { |
| 336 |
centre++; |
| 337 |
} else |
| 338 |
centre--; |
| 339 |
} |
| 340 |
|
| 341 |
if (last == LAST) |
| 342 |
*noise = centre; |
| 343 |
else { |
| 344 |
power[centre].next = LAST; |
| 345 |
power[last].next = centre; |
| 346 |
} |
| 347 |
power[centre].x = sum; |
| 348 |
power[centre].type = NOISE; |
| 349 |
last = centre; |
| 350 |
} |
| 351 |
} |
| 352 |
|
| 353 |
/**************************************************************** |
| 354 |
* |
| 355 |
* This function reduces the number of noise and tonal |
| 356 |
* component for further threshold analysis. |
| 357 |
* |
| 358 |
****************************************************************/ |
| 359 |
|
| 360 |
static void psycho_1_subsampling(mask power[HAN_SIZE], g_thres * ltg, int *tone, int *noise) |
| 361 |
{ |
| 362 |
int i, old; |
| 363 |
|
| 364 |
i = *tone; |
| 365 |
old = STOP; /* calculate tonal components for */ |
| 366 |
|
| 367 |
while ((i != LAST) && (i != STOP)) { /* reduction of spectral lines */ |
| 368 |
if (power[i].x < ltg[power[i].map].hear) { |
| 369 |
power[i].type = FALSE; |
| 370 |
power[i].x = DBMIN; |
| 371 |
if (old == STOP) |
| 372 |
*tone = power[i].next; |
| 373 |
else |
| 374 |
power[old].next = power[i].next; |
| 375 |
} else |
| 376 |
old = i; |
| 377 |
i = power[i].next; |
| 378 |
} |
| 379 |
i = *noise; |
| 380 |
old = STOP; /* calculate non-tonal components for */ |
| 381 |
while ((i != LAST) && (i != STOP)) { /* reduction of spectral lines */ |
| 382 |
if (power[i].x < ltg[power[i].map].hear) { |
| 383 |
power[i].type = FALSE; |
| 384 |
power[i].x = DBMIN; |
| 385 |
if (old == STOP) |
| 386 |
*noise = power[i].next; |
| 387 |
else |
| 388 |
power[old].next = power[i].next; |
| 389 |
} else |
| 390 |
old = i; |
| 391 |
i = power[i].next; |
| 392 |
} |
| 393 |
i = *tone; |
| 394 |
old = STOP; |
| 395 |
while ((i != LAST) && (i != STOP)) { /* if more than one */ |
| 396 |
if (power[i].next == LAST) |
| 397 |
break; /* tonal component */ |
| 398 |
if (ltg[power[power[i].next].map].bark - /* is less than .5 */ |
| 399 |
ltg[power[i].map].bark < 0.5) { /* bark, take the */ |
| 400 |
if (power[power[i].next].x > power[i].x) { /* maximum */ |
| 401 |
if (old == STOP) |
| 402 |
*tone = power[i].next; |
| 403 |
else |
| 404 |
power[old].next = power[i].next; |
| 405 |
power[i].type = FALSE; |
| 406 |
power[i].x = DBMIN; |
| 407 |
i = power[i].next; |
| 408 |
} else { |
| 409 |
power[power[i].next].type = FALSE; |
| 410 |
power[power[i].next].x = DBMIN; |
| 411 |
power[i].next = power[power[i].next].next; |
| 412 |
old = i; |
| 413 |
} |
| 414 |
} else { |
| 415 |
old = i; |
| 416 |
i = power[i].next; |
| 417 |
} |
| 418 |
} |
| 419 |
} |
| 420 |
|
| 421 |
/**************************************************************** |
| 422 |
* |
| 423 |
* This function calculates the individual threshold and |
| 424 |
* sum with the quiet threshold to find the global threshold. |
| 425 |
* |
| 426 |
****************************************************************/ |
| 427 |
|
| 428 |
/* mainly just changed the way range checking was done MFC Nov 1999 */ |
| 429 |
static void psycho_1_threshold(psycho_1_mem * mem, int *tone, int *noise, int bit_rate) |
| 430 |
{ |
| 431 |
int sub_size = mem->sub_size; |
| 432 |
mask *power = mem->power; |
| 433 |
g_thres *ltg = mem->ltg; |
| 434 |
int k, t; |
| 435 |
FLOAT dz, tmps, vf; |
| 436 |
|
| 437 |
for (k = 1; k < sub_size; k++) { |
| 438 |
ltg[k].x = DBMIN; |
| 439 |
t = *tone; /* calculate individual masking threshold for */ |
| 440 |
while ((t != LAST) && (t != STOP)) { /* components in order to find the global */ |
| 441 |
dz = ltg[k].bark - ltg[power[t].map].bark; /* distance of bark value */ |
| 442 |
if (dz >= -3.0 && dz < 8.0) { |
| 443 |
tmps = -1.525 - 0.275 * ltg[power[t].map].bark - 4.5 + power[t].x; |
| 444 |
/* masking function for lower & upper slopes */ |
| 445 |
if (dz < -1) |
| 446 |
vf = 17 * (dz + 1) - (0.4 * power[t].x + 6); |
| 447 |
else if (dz < 0) |
| 448 |
vf = (0.4 * power[t].x + 6) * dz; |
| 449 |
else if (dz < 1) |
| 450 |
vf = (-17 * dz); |
| 451 |
else |
| 452 |
vf = -(dz - 1) * (17 - 0.15 * power[t].x) - 17; |
| 453 |
ltg[k].x = add_db(mem, ltg[k].x, tmps + vf); |
| 454 |
} |
| 455 |
t = power[t].next; |
| 456 |
} |
| 457 |
|
| 458 |
t = *noise; /* calculate individual masking threshold */ |
| 459 |
while ((t != LAST) && (t != STOP)) { /* for non-tonal components to find LTG */ |
| 460 |
dz = ltg[k].bark - ltg[power[t].map].bark; /* distance of bark value */ |
| 461 |
if (dz >= -3.0 && dz < 8.0) { |
| 462 |
tmps = -1.525 - 0.175 * ltg[power[t].map].bark - 0.5 + power[t].x; |
| 463 |
/* masking function for lower & upper slopes */ |
| 464 |
if (dz < -1) |
| 465 |
vf = 17 * (dz + 1) - (0.4 * power[t].x + 6); |
| 466 |
else if (dz < 0) |
| 467 |
vf = (0.4 * power[t].x + 6) * dz; |
| 468 |
else if (dz < 1) |
| 469 |
vf = (-17 * dz); |
| 470 |
else |
| 471 |
vf = -(dz - 1) * (17 - 0.15 * power[t].x) - 17; |
| 472 |
ltg[k].x = add_db(mem, ltg[k].x, tmps + vf); |
| 473 |
} |
| 474 |
t = power[t].next; |
| 475 |
} |
| 476 |
if (bit_rate < 96) |
| 477 |
ltg[k].x = add_db(mem, ltg[k].hear, ltg[k].x); |
| 478 |
else |
| 479 |
ltg[k].x = add_db(mem, ltg[k].hear - 12.0, ltg[k].x); |
| 480 |
} |
| 481 |
|
| 482 |
} |
| 483 |
|
| 484 |
/**************************************************************** |
| 485 |
* |
| 486 |
* This function finds the minimum masking threshold and |
| 487 |
* return the value to the encoder. |
| 488 |
* |
| 489 |
****************************************************************/ |
| 490 |
|
| 491 |
static void psycho_1_minimum_mask(int sub_size, g_thres * ltg, FLOAT ltmin[SBLIMIT], int sblimit) |
| 492 |
{ |
| 493 |
FLOAT min; |
| 494 |
int i, j; |
| 495 |
|
| 496 |
j = 1; |
| 497 |
for (i = 0; i < sblimit; i++) |
| 498 |
if (j >= sub_size - 1) /* check subband limit, and */ |
| 499 |
ltmin[i] = ltg[sub_size - 1].hear; /* calculate the minimum masking */ |
| 500 |
else { /* level of LTMIN for each subband */ |
| 501 |
min = ltg[j].x; |
| 502 |
while (ltg[j].line >> 4 == i && j < sub_size) { |
| 503 |
if (min > ltg[j].x) |
| 504 |
min = ltg[j].x; |
| 505 |
j++; |
| 506 |
} |
| 507 |
ltmin[i] = min; |
| 508 |
} |
| 509 |
} |
| 510 |
|
| 511 |
/***************************************************************** |
| 512 |
* |
| 513 |
* This procedure is called in musicin to pick out the |
| 514 |
* smaller of the scalefactor or threshold. |
| 515 |
* |
| 516 |
*****************************************************************/ |
| 517 |
|
| 518 |
static void psycho_1_smr(FLOAT ltmin[SBLIMIT], FLOAT spike[SBLIMIT], FLOAT scale[SBLIMIT], |
| 519 |
int sblimit) |
| 520 |
{ |
| 521 |
int i; |
| 522 |
FLOAT max; |
| 523 |
|
| 524 |
for (i = 0; i < sblimit; i++) { /* determine the signal */ |
| 525 |
max = 20 * log10(scale[i] * 32768) - 10; /* level for each subband */ |
| 526 |
if (spike[i] > max) |
| 527 |
max = spike[i]; /* for the maximum scale */ |
| 528 |
max -= ltmin[i]; /* factors */ |
| 529 |
ltmin[i] = max; |
| 530 |
} |
| 531 |
} |
| 532 |
|
| 533 |
|
| 534 |
/* |
| 535 |
static void psycho_1_dump(mask power[HAN_SIZE], int *tone, int *noise) { |
| 536 |
int t; |
| 537 |
|
| 538 |
fprintf(stderr,"1 Ton: "); |
| 539 |
t=*tone; |
| 540 |
while (t!=LAST && t!=STOP) { |
| 541 |
fprintf(stderr,"[%i] %3.0f ",t, power[t].x); |
| 542 |
t = power[t].next; |
| 543 |
} |
| 544 |
fprintf(stderr,"\n"); |
| 545 |
|
| 546 |
fprintf(stderr,"1 Nos: "); |
| 547 |
t=*noise; |
| 548 |
while (t!=LAST && t!=STOP) { |
| 549 |
fprintf(stderr,"[%i] %3.0f ",t, power[t].x); |
| 550 |
t = power[t].next; |
| 551 |
} |
| 552 |
fprintf(stderr,"\n"); |
| 553 |
} |
| 554 |
*/ |
| 555 |
|
| 556 |
|
| 557 |
void twolame_psycho_1(twolame_options * glopts, short buffer[2][1152], FLOAT scale[2][SBLIMIT], |
| 558 |
FLOAT ltmin[2][SBLIMIT]) |
| 559 |
{ |
| 560 |
psycho_1_mem *mem; |
| 561 |
frame_header *header = &glopts->header; |
| 562 |
int nch = glopts->num_channels_out; |
| 563 |
int sblimit = glopts->sblimit; |
| 564 |
int k, i, tone = 0, noise = 0; |
| 565 |
FLOAT sample[FFT_SIZE]; |
| 566 |
FLOAT spike[2][SBLIMIT]; |
| 567 |
FLOAT *fft_buf[2]; |
| 568 |
FLOAT energy[FFT_SIZE]; |
| 569 |
|
| 570 |
/* call functions for critical boundaries, freq. */ |
| 571 |
if (!glopts->p1mem) { /* bands, bark values, and mapping */ |
| 572 |
mem = (psycho_1_mem *) TWOLAME_MALLOC(sizeof(psycho_1_mem)); |
| 573 |
|
| 574 |
mem->power = (mask_ptr) TWOLAME_MALLOC(sizeof(mask) * HAN_SIZE); |
| 575 |
if (header->version == TWOLAME_MPEG1) { |
| 576 |
mem->cbound = |
| 577 |
psycho_1_read_cbound(header->lay, header->samplerate_idx, &mem->crit_band); |
| 578 |
psycho_1_read_freq_band(&mem->ltg, header->lay, header->samplerate_idx, &mem->sub_size); |
| 579 |
} else { |
| 580 |
mem->cbound = |
| 581 |
psycho_1_read_cbound(header->lay, header->samplerate_idx + 4, &mem->crit_band); |
| 582 |
psycho_1_read_freq_band(&mem->ltg, header->lay, header->samplerate_idx + 4, |
| 583 |
&mem->sub_size); |
| 584 |
} |
| 585 |
psycho_1_make_map(mem->sub_size, mem->power, mem->ltg); |
| 586 |
for (i = 0; i < 1408; i++) |
| 587 |
mem->fft_buf[0][i] = mem->fft_buf[1][i] = 0; |
| 588 |
|
| 589 |
psycho_1_init_add_db(mem); /* create the add_db table */ |
| 590 |
|
| 591 |
mem->off[0] = 256; |
| 592 |
mem->off[1] = 256; |
| 593 |
|
| 594 |
glopts->p1mem = mem; |
| 595 |
} |
| 596 |
{ |
| 597 |
mem = glopts->p1mem; |
| 598 |
|
| 599 |
fft_buf[0] = mem->fft_buf[0]; |
| 600 |
fft_buf[1] = mem->fft_buf[1]; |
| 601 |
} |
| 602 |
|
| 603 |
|
| 604 |
for (k = 0; k < nch; k++) { |
| 605 |
/* check pcm input for 3 blocks of 384 samples */ |
| 606 |
/* sami's speedup, added in 02j saves about 4% overall during an encode */ |
| 607 |
int ok = mem->off[k] % 1408; |
| 608 |
for (i = 0; i < 1152; i++) { |
| 609 |
fft_buf[k][ok++] = (FLOAT) buffer[k][i] / SCALE; |
| 610 |
if (ok >= 1408) |
| 611 |
ok = 0; |
| 612 |
} |
| 613 |
ok = (mem->off[k] + 1216) % 1408; |
| 614 |
for (i = 0; i < FFT_SIZE; i++) { |
| 615 |
sample[i] = fft_buf[k][ok++]; |
| 616 |
if (ok >= 1408) |
| 617 |
ok = 0; |
| 618 |
} |
| 619 |
mem->off[k] += 1152; |
| 620 |
mem->off[k] %= 1408; |
| 621 |
|
| 622 |
psycho_1_hann_fft_pickmax(sample, mem->power, &spike[k][0], energy); |
| 623 |
psycho_1_tonal_label(mem, &tone); |
| 624 |
psycho_1_noise_label(mem, &noise, energy); |
| 625 |
// psycho_1_dump(power, &tone, &noise) ; |
| 626 |
psycho_1_subsampling(mem->power, mem->ltg, &tone, &noise); |
| 627 |
psycho_1_threshold(mem, &tone, &noise, glopts->bitrate / nch); |
| 628 |
psycho_1_minimum_mask(mem->sub_size, mem->ltg, <min[k][0], sblimit); |
| 629 |
psycho_1_smr(<min[k][0], &spike[k][0], &scale[k][0], sblimit); |
| 630 |
} |
| 631 |
|
| 632 |
} |
| 633 |
|
| 634 |
void twolame_psycho_1_deinit(psycho_1_mem ** mem) |
| 635 |
{ |
| 636 |
|
| 637 |
if (mem == NULL || *mem == NULL) |
| 638 |
return; |
| 639 |
|
| 640 |
TWOLAME_FREE((*mem)->cbound); |
| 641 |
TWOLAME_FREE((*mem)->ltg); |
| 642 |
TWOLAME_FREE((*mem)->power); |
| 643 |
TWOLAME_FREE((*mem)); |
| 644 |
} |
| 645 |
|
| 646 |
|
| 647 |
// vim:ts=4:sw=4:nowrap: |