1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
|
// phasevocoder.C
/******************************************************************************
*
* MiXViews - an X window system based sound & data editor/processor
*
* Copyright (c) 1993, 1994 Regents of the University of California
*
* Author: Douglas Scott
* Date: December 13, 1994
*
* Permission to use, copy and modify this software and its documentation
* for research and/or educational purposes and without fee is hereby granted,
* provided that the above copyright notice appear in all copies and that
* both that copyright notice and this permission notice appear in
* supporting documentation. The author reserves the right to distribute this
* software and its documentation. The University of California and the author
* make no representations about the suitability of this software for any
* purpose, and in no event shall University of California be liable for any
* damage, loss of data, or profits resulting from its use.
* It is provided "as is" without express or implied warranty.
*
*****************************************************************************/
#ifdef __GNUG__
#pragma implementation
#endif
#include "application.h"
#include "framedata.h"
#include "envelope.h"
#include "localdefs.h"
#include "phasevocoder.h"
#include "pipeaction.h"
extern "C" {
int fft_(float*, float*, int*, int*, int*, int*);
int reals_(float*, float*, int*, int*);
int kaiser_(int*, float*, int*, int*, float*);
}
PhaseVocoder::Info::Info(double sampRate, float scale, Mode mode,
int N, int F, int M, int D,
float T, float P, boolean useKaiser)
: fftSize(N), fundFreq(F),
inputFrameSize(M), inputFrameOffset(D),
outputFrameSize(0), outputFrameOffset(0),
pchScaleFactor(P), timeScaleFactor(T), warp(0),
firstBand(0), lastBand(0), K(useKaiser),
inputScalingFactor(scale), samplingRate(sampRate),
runMode(mode) {}
PhaseVocoder::Info::Info(double samprate, float scale,
int N, int F, int D, int L, int I,
float T, int i, int j, float warpFactor,
boolean useKaiser)
: fftSize(N), fundFreq(F),
inputFrameSize(0), inputFrameOffset(D),
outputFrameSize(L), outputFrameOffset(I),
pchScaleFactor(1.0), timeScaleFactor(T),
warp(warpFactor), firstBand(i), lastBand(j), K(useKaiser),
inputScalingFactor(scale), samplingRate(samprate),
runMode(PhaseVocoder::Synthesis) {}
float PhaseVocoder::Pi = 4. * atan(1.);
float PhaseVocoder::HalfPi = 2. * atan(1.);
float PhaseVocoder::TwoPi = 8. * atan(1.);
float PhaseVocoder::beta = 6.8; // for Kaiser window
// ctor for analysis only
PhaseVocoder::PhaseVocoder(Info& info) : I(info), initialized(false) {
zeroPointers();
setDefaults();
setUpLog();
switch(I.runMode) {
case Spectrum:
E = 1;
break;
case Magnitudes:
X = 1;
break;
default:
break;
}
if(checkAndSetValues())
initialize();
}
void
PhaseVocoder::setDefaults() {
E = X = 0;
analWinLen = synWinLen = 0;
startingOffset = 0;
verbose = true;
ftot = 0.; // scale factor for calculating statistics
inputFrameEven = 0; // flag for even inputFrameSize
outputFrameEven = 0; // flag for even outputFrameSize
C = 0;
bandsLimited = 0;
}
PhaseVocoder::~PhaseVocoder() {
// printToLog();
Resource::unref(timeScaleEnvelope);
delete [] output,
delete [] anal,
delete [] syn,
delete [] analWindowBuf,
delete [] synWindowBuf,
delete [] maxAmp,
delete [] avgAmp,
delete [] avgFrq,
delete [] env,
delete [] oldInPhase,
delete [] oldOutPhase;
}
// zero these to avoid deleting uninitialized pointers if in error condition
void
PhaseVocoder::zeroPointers() {
timeScaleEnvelope = nil;
output = nil;
anal = nil;
syn = nil;
analWindowBuf = nil;
synWindowBuf = nil;
maxAmp = nil;
avgAmp = nil;
avgFrq = nil;
env = nil;
oldInPhase = nil;
oldOutPhase = nil;
}
boolean
PhaseVocoder::isGood() { return initialized; }
void
PhaseVocoder::setUpLog() {
if (verbose)
fp = stderr; // write out to stderr for now
// fp = fopen("pvoc.log","w");
}
void
PhaseVocoder::printToLog() {
if (verbose && isGood() && analyzing()) { // dont do this if never configured!
if (ftot != 0.)
ftot = 1. / ftot;
fprintf(fp,"\n i band max amp ");
fprintf(fp,"avg amp avg frq\n\n");
int F2 = I.fundFreq / 2;
for (int i = 0; i <= fftPoints; i++) fprintf(fp,
"%4d %5d - %5d %8.5f %8.5f %8.1f\n",
(i<<1)+1, i*I.fundFreq-F2, i*I.fundFreq+F2, *(maxAmp+i),
*(avgAmp+i)*ftot, *(avgFrq+i)*ftot);
fprintf(fp,"\n");
}
}
/*
user friendliness: Try to catch bad parameter specifications and correct
them if possible. The basic idea is that we step through the
signal applying an inputFrameSize point window and taking an fftSize
point FFT at a spacing of inputFrameOffset samples. The synthesis consists
of taking an invFftSize point inverse FFT, applying an outputFrameSize
point window, and overlap-adding at a spacing of outputFrameOffset samples.
And in the middle, we linearly rescale the phase by the time-scale
expansion factor timeScaleFactor = outputFrameOffset / inputFrameOffset.
The default is outputFrameSize = inputFrameSize = fftSize and
outputFrameOffset = inputFrameOffset, which gives an analysis-
synthesis identity. NOTE: the actual timeScaleFactor will not necessarily
equal the specified T - if this is important, then specify inputFrameOffset
and outputFrameOffset directly. NOTE: pitch transposition is performed by
changing the size of the inverse FFT relative to the forward
FFT - this is a cheap trick which works best on octave
transpositions. NOTE: the spectral envelope estimation and
warping is relatively crude.
*/
boolean
PhaseVocoder::checkAndSetValues() {
char errmsg[128];
errmsg[0] = '\0';
if ((I.fftSize != 0) && (I.fundFreq != 0))
sprintf(errmsg,"Don't specify both fftSize and fundFreq.");
if ((I.fftSize == 0) && (I.fundFreq == 0))
I.fftSize = 256;
else if (I.fundFreq != 0)
I.fftSize = int(I.samplingRate / I.fundFreq);
if ((I.fftSize%2) != 0)
I.fftSize += 1; /* even values usually run faster */
fftPoints = I.fftSize / 2;
if (fftPoints > 16384) {
sprintf(errmsg,"FFT size (%d) too large.", fftPoints);
goto error;
}
I.fundFreq = int(I.samplingRate / I.fftSize);
if (I.inputFrameSize == 0)
I.inputFrameSize = I.fftSize;
if ((I.inputFrameSize%2) == 0)
inputFrameEven = 1;
if (I.outputFrameSize == 0)
I.outputFrameSize = I.inputFrameSize;
if ((I.outputFrameSize%2) == 0)
outputFrameEven = 1;
if (I.inputFrameSize < 7) {
sprintf(errmsg,"Input frame size (%d) is too small.", I.inputFrameSize);
goto error;
}
obuflen = 4 * I.outputFrameSize;
if (I.lastBand == 0) I.lastBand = fftPoints;
if (I.lastBand > fftPoints) I.lastBand = fftPoints;
if (I.firstBand < 0) I.firstBand = 0;
bandsLimited = (I.firstBand != 0 || I.lastBand != fftPoints);
if ((I.pchScaleFactor != 1.) && (I.timeScaleFactor != 1.))
sprintf(errmsg, "Don't specify both time and pitch scaling -- using pitch scaling.");
if ((I.pchScaleFactor != 1.) && variableTimeScaling()) {
sprintf(errmsg, "Sorry, no pitch scaling with time-varying time scaling.");
goto error;
}
if (I.pchScaleFactor != 1.)
I.timeScaleFactor = I.pchScaleFactor; /* pitch change is time change plus resamp */
if (I.timeScaleFactor <= 0.){
sprintf(errmsg,"Invalid time scale factor (%f).", I.timeScaleFactor);
goto error;
}
if (I.inputFrameOffset == 0)
if (I.timeScaleFactor > 1.)
I.inputFrameOffset = int(I.inputFrameSize / (8.0 * I.timeScaleFactor));
else
I.inputFrameOffset = int(I.inputFrameSize / 8.0);
if (I.inputFrameOffset == 0){
sprintf(errmsg,"Time scale factor greater than inputFrameSize/8.");
I.inputFrameOffset = 1;
}
// if doing resynthesis, remultiply outputFrameOffset by time factor
// to return to "unity" value
if (I.outputFrameOffset == 0)
I.outputFrameOffset = synthesizing() ?
int(I.inputFrameOffset / I.timeScaleFactor)
: int(I.timeScaleFactor * I.inputFrameOffset);
if (I.outputFrameOffset == 0){
sprintf(errmsg, "time factor * offset (or pitch factor * offset) < 1. Increase input frame size.");
goto error;
}
if (variableTimeScaling()) {
// original timeScaleFactor was maximum - used to set outputFrameOffset
I.timeScaleFactor = timeScaleEnvelope->get(0);
I.inputFrameOffset = int(I.outputFrameOffset / I.timeScaleFactor);
if (I.inputFrameOffset < 1) {
char msg[128];
sprintf(msg, "Cannot expand by time scale factor %f.",
I.timeScaleFactor);
Application::alert(msg, "Setting to factor to maximum value.");
I.inputFrameOffset = 1;
}
if (I.warp != 0.)
I.warp = I.timeScaleFactor; // warp varies with timeScaleFactor
}
if(variableTimeScaling() || !synthesizing())
I.timeScaleFactor = (float(I.outputFrameOffset) / I.inputFrameOffset);
if (I.pchScaleFactor != 1.)
I.pchScaleFactor = I.timeScaleFactor;
// synth transform will be invFftSize points
invFftSize = int(I.fftSize / I.pchScaleFactor);
if ((invFftSize % 2) != 0)
invFftSize += 1;
invFftPoints = invFftSize / 2;
/* ideally, fftSize/invFftSize = I.outputFrameOffset/I.inputFrameOffset
= pitch change
*/
I.pchScaleFactor = (float(I.fftSize) / invFftSize);
if (I.warp == -1.)
I.warp = I.pchScaleFactor;
if ((E == 1) && (I.warp == 0.))
I.warp = 1.;
if ((I.pchScaleFactor != 1.) && (I.pchScaleFactor != I.timeScaleFactor))
sprintf(errmsg,"pchScaleFactor (%f) not equal to timeScaleFactor (%f).",I.pchScaleFactor,I.timeScaleFactor);
scaledOutputOffset = int(I.outputFrameOffset / I.pchScaleFactor);
if (verbose) {
fprintf(fp,"\nfftSize: %d I.inputFrameSize: %d outputFrameSize: %d",I.fftSize,I.inputFrameSize,I.outputFrameSize);
fprintf(fp," inputFrameOffset: %d outputFrameOffset: %d fundFreq: %d",I.inputFrameOffset,I.outputFrameOffset,I.fundFreq);
fprintf(fp," sampleRate: %d pchScaleFactor: %5.2f timeScaleFactor: %5.2f\n",int(I.samplingRate),I.pchScaleFactor,I.timeScaleFactor);
if (bandsLimited)
fprintf(fp,"C: %d i: %d j: %d\n",C,I.firstBand,I.lastBand);
if (I.K)
fprintf(fp,"---Kaiser Window---\n");
}
if(strlen(errmsg))
Application::alert("Warning:", errmsg);
return true;
error:
Application::alert(errmsg);
return false;
}
void
PhaseVocoder::initialize() {
createAnalysisWindow();
createSynthesisWindow();
createBuffers();
reset();
initialized = true;
}
// initialization: input time starts negative so that the rightmost
// edge of the analysis filter just catches the first non-zero
// input samples; output time is always timeScaleFactor times input time.
void
PhaseVocoder::reset() {
outCount = 0;
float inputFrameRate = I.samplingRate / I.inputFrameOffset;
float outputFrameRate = I.samplingRate / I.outputFrameOffset;
RoverTwoPi = inputFrameRate / TwoPi;
TwoPioverR = TwoPi / outputFrameRate;
// input time (in samples, starts negative)
// inSamp is -1 * integer number of offsets per half-window
inSamp = -(analWinLen/I.inputFrameOffset) * I.inputFrameOffset;
// output time (in samps, also starts negative)
outSamp = int(I.timeScaleFactor/I.pchScaleFactor * inSamp);
sampsIn = analWinLen + inSamp + 1; // number of new inputs to read
startingOffset = sampsIn; // cached here
sampsOut = 0;
}
// set up analysis window: The window is assumed to be symmetric
// with inputFrameSize total points. After the initial memory allocation,
// analWindow always points to the midpoint of the window
// (or one half sample to the right, if inputFrameSize is even);
// analWinLen is half the true window length (rounded down). Any low pass
// window will work; a Hamming window is generally fine,
// but a Kaiser is also available. If the window duration is
// MXINT32er than the transform (inputFrameSize > fftSize), then the
// window is multiplied by a sin(x)/x function to meet the condition:
// analWindow[Ni] = 0 for i != 0. In either case, the
// window is renormalized so that the phase vocoder amplitude
// estimates are properly scaled.
void
PhaseVocoder::createAnalysisWindow() {
analWindowBuf = new float[I.inputFrameSize+inputFrameEven];
bzero(analWindowBuf, (I.inputFrameSize+inputFrameEven) * sizeof(float));
analWindow = analWindowBuf + (analWinLen = I.inputFrameSize/2);
int one = 1;
int i = 0;
if (I.K)
kaiser_(&I.inputFrameSize,analWindow,&analWinLen,&one,&beta);
else
hamming(analWindow,analWinLen,inputFrameEven);
for (i = 1; i <= analWinLen; i++)
*(analWindow - i) = *(analWindow + i - inputFrameEven);
if (I.inputFrameSize > I.fftSize) {
if (inputFrameEven)
*analWindow *= I.fftSize * sin(HalfPi/I.fftSize) / HalfPi;
for (i = 1; i <= analWinLen; i++)
*(analWindow + i) *=
I.fftSize * sin(Pi*(i+.5*inputFrameEven)/I.fftSize) /
(Pi*(i+.5*inputFrameEven));
for (i = 1; i <= analWinLen; i++)
*(analWindow - i) = *(analWindow + i - inputFrameEven);
}
/* normalize window for unity gain across unmodified
analysis-synthesis procedure */
sum = 0.;
for (i = -analWinLen; i < analWinLen; i++)
sum += *(analWindow + i);
sum = 2. / sum; // factor of 2 comes in later in trig identity
if(analyzing())
sum *= I.inputScalingFactor; // input scaled by 1/(2^16) if short integer
for (i = -analWinLen; i < analWinLen; i++)
*(analWindow + i) *= sum;
}
// set up synthesis window: For the minimal mean-square-error
// formulation (valid for fftSize >= inputFrameSize), the synthesis window
// is identical to the analysis window (except for a
// scale factor), and both are even in length. If
// fftSize < inputFrameSize, then an interpolating synthesis window is used.
void
PhaseVocoder::createSynthesisWindow() {
synWindowBuf = new float[I.outputFrameSize+outputFrameEven];
bzero(synWindowBuf, (I.outputFrameSize+outputFrameEven) * sizeof(float));
synWindow = synWindowBuf + (synWinLen = I.outputFrameSize/2);
int one = 1;
int i = 0;
if (I.inputFrameSize <= I.fftSize) {
if (I.K)
kaiser_(&I.inputFrameSize,synWindow,&synWinLen,&one,&beta);
else
hamming(synWindow,synWinLen,outputFrameEven);
for (i = 1; i <= synWinLen; i++)
*(synWindow - i) = *(synWindow + i - outputFrameEven);
for (i = -synWinLen; i < synWinLen; i++)
*(synWindow + i) *= sum;
// normalize window for unity gain across unmodified
// analysis-synthesis procedure
sum = 0.;
for (i = -synWinLen; i <= synWinLen; i+=I.outputFrameOffset)
sum += *(synWindow + i) * *(synWindow + i);
}
else {
hamming(synWindow,synWinLen,outputFrameEven);
for (i = 1; i <= synWinLen; i++)
*(synWindow - i) = *(synWindow + i - outputFrameEven);
if (outputFrameEven)
*synWindow *= scaledOutputOffset * sin(HalfPi/scaledOutputOffset) /
HalfPi;
for (i = 1; i <= synWinLen; i++)
*(synWindow + i) *= scaledOutputOffset *
sin(Pi*(i+.5*outputFrameEven) / scaledOutputOffset) /
(Pi*(i+.5*outputFrameEven));
for (i = 1; i <= synWinLen; i++)
*(synWindow - i) = *(synWindow + i - outputFrameEven);
}
sum = 1. / sum;
if(synthesizing())
sum *= I.inputScalingFactor; // output mult. by (2^16) if short integer
for (i = -synWinLen; i < synWinLen; i++)
*(synWindow + i) *= sum;
}
void
PhaseVocoder::hamming(float *win, int winLen, int even) {
int i;
float ftmp = Pi/winLen;
if (even) {
for (i=0; i<winLen; i++)
*(win+i) = .54 + .46 * cos(ftmp*(i+.5));
*(win+winLen) = 0.;
}
else {
*(win) = 1.;
for (i=1; i<=winLen; i++)
*(win+i) = .54 + .46 * cos(ftmp*i);
}
}
void
PhaseVocoder::createBuffers() {
/* set up output buffer: nextOut always points to the next word
to be shifted out. The shift is simulated by writing the
value to the standard output and then setting that word
of the buffer to zero. When nextOut reaches the end of
the buffer, it jumps back to the beginning. */
nextOut = output = new float[obuflen];
bzero((char*) output, sizeof(float) * obuflen);
/* set up analysis buffer for (fftSize/2 + 1) channels: The input is real,
so the other channels are redundant. oldInPhase is used
in the conversion to remember the previous phase when
calculating phase difference between successive samples. */
anal = new float[analysisChannels()];
// allocate and zero these
int fb = freqBands();
oldInPhase = new float[fb];
bzero((char*) oldInPhase, sizeof(float) * fb);
maxAmp = new float[fb];
bzero((char*) maxAmp, sizeof(float) * fb);
avgAmp = new float[fb];
bzero((char*) avgAmp, sizeof(float) * fb);
avgFrq = new float[fb];
bzero((char*) avgFrq, sizeof(float) * fb);
env = new float[fb];
bzero((char*) env, sizeof(float) * fb);
/* set up synthesis buffer for (fftSize/2 + 1) channels: (This is included
only for clarity.) oldOutPhase is used in the re-
conversion to accumulate angle differences (actually angle
difference per second). */
syn = new float[invFftSize+2];
oldOutPhase = new float[invFftPoints + 1];
bzero((char*) oldOutPhase, sizeof(float) * (invFftPoints + 1));
}
void
PhaseVocoder::updateAmps() {
ftot++;
for (int i = 0; i <= fftPoints; i++){
if (*(anal+(i<<1)) > *(maxAmp+i))
*(maxAmp+i) = *(anal+(i<<1));
*(avgAmp + i) += *(anal + (i<<1));
*(avgFrq + i) += *(anal + (i<<1) + 1);
}
}
int
PhaseVocoder::runAnalysis(double* in, Data* frame) {
applyInputWindow(in);
analyze();
convertToReal();
detectAndWarpEnvelope();
// if (verbose) updateAmps();
outputAnalysis(frame);
getTimeScaleFactor();
calculateOffsetsAndIncrement();
return true;
}
/*
analysis: The analysis subroutine computes the complex output at
time n of (fftSize/2 + 1) of the phase vocoder channels. It operates
on input samples (n - analWinLen) thru (n + analWinLen) and
expects to find these in input[0 - 2*analWinLen)].
This subroutine expects analWindow to point to the center of a
symmetric window of length (2 * analWinLen +1). It is the
responsibility of the main program to ensure that these values
are correct! The results are returned in anal as succesive
pairs of real and imaginary values for the lowest (fftSize/2 + 1)
channels. The subroutines fft and reals together implement
one efficient FFT call for a real input sequence.
*/
void
PhaseVocoder::applyInputWindow(double* input) {
int i = 0;
for (i = 0; i < analysisChannels(); i++) // initialize
*(anal + i) = 0.;
int k = inSamp - analWinLen - 1; // time shift
while (k < 0) // get rotated pointer
k += I.fftSize;
k = k % I.fftSize;
for (i = -analWinLen; i <= analWinLen; i++) {
if (++k >= I.fftSize)
k = 0;
*(anal + k) += *(analWindow + i) * *input++;
}
}
// extern void rfft(float*, int, int);
void
PhaseVocoder::analyze() {
register float* banal = anal + 1;
int one = 1;
int negTwo = -2;
fft_(anal,banal,&one,&fftPoints,&one,&negTwo);
reals_(anal,banal,&fftPoints,&negTwo);
// rfft(anal, fftPoints, true);
}
void
PhaseVocoder::convertToReal() {
/*
conversion: The real and imaginary values in anal are converted to
magnitude and angle-difference-per-second (assuming an
intermediate sampling rate of inputFrameRate) and are returned in anal.
*/
register float* realptr = anal;
register float* imagptr = anal + 1;
register float* phaseptr = oldInPhase;
for (int i = 0; i <= fftPoints;
i++, realptr += 2, imagptr += 2, phaseptr++) {
float real = *realptr;
float imag = *imagptr;
float angleDif = 0;
*realptr = hypot(real,imag);
if (*realptr == 0.)
angleDif = 0.;
else {
float phase = 0;
angleDif = (phase = atan2(imag,real)) - *phaseptr;
*phaseptr = phase;
}
// unwrap phase differences
while(angleDif > Pi)
angleDif -= TwoPi;
while(angleDif < -Pi)
angleDif += TwoPi;
// convert each phase difference to Hz
*imagptr = angleDif * RoverTwoPi + (i * float(I.fundFreq));
}
}
void
PhaseVocoder::outputAnalysis(Data* frame) {
if(E)
frame->setFrame(env, fftPoints); // spectral envelope
else if(X)
for (int i=0; i <= fftPoints; i++) // magnitudes only
frame->set(*(anal + (i<<1)), 0, i);
else
frame->setFrame(anal, analysisChannels()); // pvoc analysis data
}
void
PhaseVocoder::detectAndWarpEnvelope() {
/*
spectral envelope detection: this is a very crude peak picking algorithm
which is used to detect and pre-warp the spectral envelope so that
pitch transposition can be performed without altering timbre.
The basic idea is to disallow large negative slopes between
successive values of magnitude vs. frequency.
*/
if (I.warp != 0.){
float lastmag = *anal;
float mag = *(anal + 2);
float nextmag = 0;
float eps = -64. / I.fftSize;
float pkOld = lastmag;
*env = pkOld;
int pkcnt = 1;
float slope = 0;
int i = 0;
for (i = 1; i <= fftPoints; i++) { // step thru spectrum
if (i<fftPoints)
nextmag = *(anal + (i<<1) + 2);
else nextmag = 0.;
if (pkOld != 0.)
slope = (float(mag - pkOld)/(pkOld * pkcnt));
else
slope = -10.;
// look for peaks
if ((mag>=lastmag)&&(mag>nextmag)&&(slope>eps)) {
*(env + i) = mag;
pkcnt--;
for (int j = 1; j <= pkcnt; j++)
//*(env + i - pkcnt + j - 1) = pkOld * (1. + slope * j);
env[i - pkcnt + j - 1] = pkOld * (1. + slope * j);
pkOld = mag;
pkcnt = 1;
}
else pkcnt++; // not a peak
lastmag = mag;
mag = nextmag;
}
if (pkcnt > 1) { // get final peak
mag = *(anal + I.fftSize);
slope = (float(mag - pkOld) / pkcnt);
*(env + fftPoints) = mag;
pkcnt--;
for (int j = 1; j <= pkcnt; j++)
//*(env + fftPoints - pkcnt + j - 1) = pkOld + slope * j;
env[fftPoints - pkcnt + j - 1] = pkOld + slope * j;
}
for (i = 0; i <= fftPoints; i++) { // warp spectral env.
int j = int(i * I.warp);
if ((j <= fftPoints) && (*(env + i) != 0.))
*(anal + (i<<1)) *= *(env + j) / *(env + i);
else
*(anal + (i<<1)) = 0.;
}
}
}
int
PhaseVocoder::runSynthesis(Data* frame, InPipeAction* pipe) {
BUG("PhaseVocoder::runSynthesis");
loadAnalysis(frame);
limitBands();
detectAndWarpEnvelope(); // try this here!!
convertFromReal();
synthesize();
applyOutputWindow();
int status = shiftOut(pipe);
getTimeScaleFactor();
calculateOffsetsAndIncrement();
return status;
}
void
PhaseVocoder::loadAnalysis(Data* frame) {
BUG("PhaseVocoder::loadAnalysis");
frame->getFrame(anal, analysisChannels());
}
void
PhaseVocoder::limitBands() {
/* resynthesize only selected channels */
if (bandsLimited){
int i;
for (i = 0; i < I.firstBand; i++)
*(anal+(i<<1)) = 0.;
for (i = I.lastBand+1; i <= fftPoints; i++)
*(anal+(i<<1)) = 0.;
if (C == 1)
for (i = I.firstBand; i <= I.lastBand; i++)
if (i%2 == 0)
*(anal+(i<<1)) = 0.;
if (C == 2)
for (i = I.firstBand; i <= I.lastBand; i++)
if (i%2 != 0)
*(anal+(i<<1)) = 0.;
}
}
void
PhaseVocoder::convertFromReal() {
/*
reconversion: The magnitude and angle-difference-per-second in syn
(assuming an intermediate sampling rate of outputFrameRate) are
converted to real and imaginary values and are returned in syn.
This automatically incorporates the proper phase scaling for
time modifications.
*/
int i = 0;
if (invFftSize <= I.fftSize){
for (i = 0; i < invFftSize+2; i++)
*(syn+i) = *(anal+i);
}
else {
for (i = 0; i <= I.fftSize+1; i++)
*(syn+i) = *(anal+i);
for (i = analysisChannels(); i < invFftSize+2; i++)
*(syn+i) = 0.;
}
register float* i0 = syn;
register float* i1 = syn + 1;
for (i = 0; i <= invFftPoints; i++, i0 += 2, i1 += 2) {
float mag = *i0;
*(oldOutPhase + i) += *i1 - ( i * float(I.fundFreq));
float phase = *(oldOutPhase + i) * TwoPioverR;
*i0 = mag * cos(phase);
*i1 = mag * sin(phase);
}
// scale values by pitch factor if present
if (I.pchScaleFactor != 1.) {
float Pinv = 1.0 / I.pchScaleFactor;
for (i = 0; i < invFftSize+2; i++)
*(syn+i) *= Pinv;
}
}
/*
synthesis: The synthesis subroutine uses the Weighted Overlap-Add
technique to reconstruct the time-domain signal. The (fftSize/2 + 1)
phase vocoder channel outputs at time n are inverse Fourier
transformed, windowed, and added into the output array. The
subroutine thinks of output as a shift register in which
locations are referenced modulo obuflen. Therefore, the main
program must take care to zero each location which it "shifts"
out (to standard output). The subroutines reals and fft
together perform an efficient inverse FFT.
*/
void
PhaseVocoder::synthesize() {
register float* bsyn = syn + 1;
int one = 1;
int two = 2;
reals_(syn,bsyn,&invFftPoints,&two);
fft_(syn,bsyn,&one,&invFftPoints,&one,&two);
}
void
PhaseVocoder::applyOutputWindow() {
BUG("PhaseVocoder::applyOutputWindow");
// set up pointers to current locations in syn and output
int j = outSamp - synWinLen - 1;
while (j < 0)
j += obuflen;
j = j % obuflen;
int k = outSamp - synWinLen - 1;
while (k < 0)
k += invFftSize;
k = k % invFftSize;
// overlap-add windowed synth buffer into output array, wrapping pointers
// as required
for (int i = -synWinLen; i <= synWinLen; i++) {
if (++j >= obuflen)
j -= obuflen;
if (++k >= invFftSize)
k -= invFftSize;
*(output + j) += *(syn + k) * *(synWindow + i);
}
}
int
PhaseVocoder::shiftOut(InPipeAction* pipe) {
BUG("PhaseVocoder::shiftOut");
// shift out next sampsOut values
pipe->ref();
int keepGoing = true;
for (int i = 0; i < sampsOut && keepGoing; i++) {
keepGoing = pipe->add(*nextOut);
*(nextOut++) = 0.;
if (nextOut >= (output + obuflen))
nextOut -= obuflen;
outCount++;
}
Resource::unref(pipe);
return keepGoing;
}
void
PhaseVocoder::getTimeScaleFactor() {
// time-varying time-scaling: get linearly interpolated time values
// from envelope object
if (variableTimeScaling() && inSamp > 0) { // if window midpt. is positive
I.timeScaleFactor = timeScaleEnvelope->next();
if (I.timeScaleFactor < (8.0 * I.outputFrameOffset / I.inputFrameSize)) {
char msg[64];
sprintf(msg, "Cannot contract by time scaling %f.",
I.timeScaleFactor);
Application::inform(msg, true); // display for 1 second
I.timeScaleFactor = (8.0 * I.outputFrameOffset / (I.inputFrameSize + 1));
}
I.inputFrameOffset = int(I.outputFrameOffset / I.timeScaleFactor);
if (I.inputFrameOffset < 1) {
char msg[64];
sprintf(msg,"Cannot expand by time scaling %f.", I.timeScaleFactor);
Application::inform(msg, true); // display for 1 second
I.inputFrameOffset = 1;
}
I.timeScaleFactor = (float(I.outputFrameOffset) / I.inputFrameOffset);
float inputFrameRate = (I.samplingRate / I.inputFrameOffset);
RoverTwoPi = inputFrameRate / TwoPi;
if (I.warp != 0.)
I.warp = I.timeScaleFactor;
}
}
void
PhaseVocoder::calculateOffsetsAndIncrement() {
/* I.inputFrameOffset = some_function(inSamp); for variable time-scaling */
/* float inputFrameRate = (I.samplingRate / I.inputFrameOffset); for variable time-scaling */
/* RoverTwoPi = inputFrameRate / TwoPi; for variable time-scaling */
inSamp += I.inputFrameOffset; /* increment time */
outSamp += scaledOutputOffset;
sampsIn = I.inputFrameOffset; // always, after first time
float Ii = 0.0;
if (outSamp > (synWinLen + I.outputFrameOffset))
Ii = I.outputFrameOffset;
else if (outSamp > synWinLen)
Ii = outSamp - synWinLen;
else {
Ii = 0;
for (int i=outSamp+synWinLen; i<obuflen; i++)
if (i > 0)
*(output+i) = 0.;
}
sampsOut = int(Ii / I.pchScaleFactor);
}
int
PhaseVocoder::calculateAnalysisLength(int inputLen) {
return max(1,
1 + (inputLen + I.inputFrameSize-startingOffset) / I.inputFrameOffset
);
}
|