Audio quality improvements in PitchShift
(cherry picked from commit 29b6ed9283
)
This commit is contained in:
parent
6471d0a57a
commit
51d028fce0
|
@ -74,7 +74,7 @@
|
||||||
*
|
*
|
||||||
*****************************************************************************/
|
*****************************************************************************/
|
||||||
|
|
||||||
void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long fftFrameSize, long osamp, float sampleRate, float *indata, float *outdata,int stride) {
|
void SMBPitchShift::PitchShift(float pitchShift, int64_t numSampsToProcess, int64_t fftFrameSize, int64_t osamp, float sampleRate, float *indata, float *outdata,int stride) {
|
||||||
|
|
||||||
|
|
||||||
/*
|
/*
|
||||||
|
@ -85,19 +85,32 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
*/
|
*/
|
||||||
|
|
||||||
double magn, phase, tmp, window, real, imag;
|
double magn, phase, tmp, window, real, imag;
|
||||||
double freqPerBin, expct;
|
double freqPerBin, expct, reciprocalFftFrameSize;
|
||||||
long i,k, qpd, index, inFifoLatency, stepSize, fftFrameSize2;
|
int64_t i,k, qpd, index, inFifoLatency, stepSize, fftFrameSize2;
|
||||||
|
|
||||||
/* set up some handy variables */
|
/* set up some handy variables */
|
||||||
fftFrameSize2 = fftFrameSize/2;
|
fftFrameSize2 = fftFrameSize/2;
|
||||||
|
reciprocalFftFrameSize = 1./fftFrameSize;
|
||||||
stepSize = fftFrameSize/osamp;
|
stepSize = fftFrameSize/osamp;
|
||||||
freqPerBin = sampleRate/(double)fftFrameSize;
|
freqPerBin = reciprocalFftFrameSize * sampleRate;
|
||||||
expct = 2.*Math_PI*(double)stepSize/(double)fftFrameSize;
|
expct = Math_TAU * reciprocalFftFrameSize * stepSize;
|
||||||
inFifoLatency = fftFrameSize-stepSize;
|
inFifoLatency = fftFrameSize-stepSize;
|
||||||
if (gRover == 0) { gRover = inFifoLatency;
|
if (gRover == 0) {
|
||||||
}
|
gRover = inFifoLatency;
|
||||||
|
}
|
||||||
|
|
||||||
/* initialize our static arrays */
|
// If pitchShift changes clear arrays to prevent some artifacts and quality loss.
|
||||||
|
if (lastPitchShift != pitchShift) {
|
||||||
|
lastPitchShift = pitchShift;
|
||||||
|
memset(gInFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
||||||
|
memset(gOutFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
||||||
|
memset(gFFTworksp, 0, 2 * MAX_FRAME_LENGTH * sizeof(double));
|
||||||
|
memset(gLastPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(double));
|
||||||
|
memset(gSumPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(double));
|
||||||
|
memset(gOutputAccum, 0, 2 * MAX_FRAME_LENGTH * sizeof(double));
|
||||||
|
memset(gAnaFreq, 0, MAX_FRAME_LENGTH * sizeof(double));
|
||||||
|
memset(gAnaMagn, 0, MAX_FRAME_LENGTH * sizeof(double));
|
||||||
|
}
|
||||||
|
|
||||||
/* main processing loop */
|
/* main processing loop */
|
||||||
for (i = 0; i < numSampsToProcess; i++){
|
for (i = 0; i < numSampsToProcess; i++){
|
||||||
|
@ -113,7 +126,7 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
|
|
||||||
/* do windowing and re,im interleave */
|
/* do windowing and re,im interleave */
|
||||||
for (k = 0; k < fftFrameSize;k++) {
|
for (k = 0; k < fftFrameSize;k++) {
|
||||||
window = -.5*cos(2.*Math_PI*(double)k/(double)fftFrameSize)+.5;
|
window = -.5*cos(Math_TAU * reciprocalFftFrameSize * k)+.5;
|
||||||
gFFTworksp[2*k] = gInFIFO[k] * window;
|
gFFTworksp[2*k] = gInFIFO[k] * window;
|
||||||
gFFTworksp[2*k+1] = 0.;
|
gFFTworksp[2*k+1] = 0.;
|
||||||
}
|
}
|
||||||
|
@ -143,13 +156,15 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
|
|
||||||
/* map delta phase into +/- Pi interval */
|
/* map delta phase into +/- Pi interval */
|
||||||
qpd = tmp/Math_PI;
|
qpd = tmp/Math_PI;
|
||||||
if (qpd >= 0) { qpd += qpd&1;
|
if (qpd >= 0) {
|
||||||
} else { qpd -= qpd&1;
|
qpd += qpd&1;
|
||||||
}
|
} else {
|
||||||
|
qpd -= qpd&1;
|
||||||
|
}
|
||||||
tmp -= Math_PI*(double)qpd;
|
tmp -= Math_PI*(double)qpd;
|
||||||
|
|
||||||
/* get deviation from bin frequency from the +/- Pi interval */
|
/* get deviation from bin frequency from the +/- Pi interval */
|
||||||
tmp = osamp*tmp/(2.*Math_PI);
|
tmp = osamp*tmp/Math_TAU;
|
||||||
|
|
||||||
/* compute the k-th partials' true frequency */
|
/* compute the k-th partials' true frequency */
|
||||||
tmp = (double)k*freqPerBin + tmp*freqPerBin;
|
tmp = (double)k*freqPerBin + tmp*freqPerBin;
|
||||||
|
@ -162,8 +177,8 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
|
|
||||||
/* ***************** PROCESSING ******************* */
|
/* ***************** PROCESSING ******************* */
|
||||||
/* this does the actual pitch shifting */
|
/* this does the actual pitch shifting */
|
||||||
memset(gSynMagn, 0, fftFrameSize*sizeof(float));
|
memset(gSynMagn, 0, fftFrameSize*sizeof(double));
|
||||||
memset(gSynFreq, 0, fftFrameSize*sizeof(float));
|
memset(gSynFreq, 0, fftFrameSize*sizeof(double));
|
||||||
for (k = 0; k <= fftFrameSize2; k++) {
|
for (k = 0; k <= fftFrameSize2; k++) {
|
||||||
index = k*pitchShift;
|
index = k*pitchShift;
|
||||||
if (index <= fftFrameSize2) {
|
if (index <= fftFrameSize2) {
|
||||||
|
@ -187,7 +202,7 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
tmp /= freqPerBin;
|
tmp /= freqPerBin;
|
||||||
|
|
||||||
/* take osamp into account */
|
/* take osamp into account */
|
||||||
tmp = 2.*Math_PI*tmp/osamp;
|
tmp = Math_TAU*tmp/osamp;
|
||||||
|
|
||||||
/* add the overlap phase advance back in */
|
/* add the overlap phase advance back in */
|
||||||
tmp += (double)k*expct;
|
tmp += (double)k*expct;
|
||||||
|
@ -202,35 +217,35 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
|
||||||
}
|
}
|
||||||
|
|
||||||
/* zero negative frequencies */
|
/* zero negative frequencies */
|
||||||
for (k = fftFrameSize+2; k < 2*fftFrameSize; k++) { gFFTworksp[k] = 0.;
|
for (k = fftFrameSize+2; k < 2*MAX_FRAME_LENGTH; k++) {
|
||||||
}
|
gFFTworksp[k] = 0.;
|
||||||
|
}
|
||||||
|
|
||||||
/* do inverse transform */
|
/* do inverse transform */
|
||||||
smbFft(gFFTworksp, fftFrameSize, 1);
|
smbFft(gFFTworksp, fftFrameSize, 1);
|
||||||
|
|
||||||
/* do windowing and add to output accumulator */
|
/* do windowing and add to output accumulator */
|
||||||
for(k=0; k < fftFrameSize; k++) {
|
for(k=0; k < fftFrameSize; k++) {
|
||||||
window = -.5*cos(2.*Math_PI*(double)k/(double)fftFrameSize)+.5;
|
window = -.5*cos(Math_TAU * reciprocalFftFrameSize * k)+.5;
|
||||||
gOutputAccum[k] += 2.*window*gFFTworksp[2*k]/(fftFrameSize2*osamp);
|
gOutputAccum[k] += 2.*window*gFFTworksp[2*k]/(fftFrameSize2*osamp);
|
||||||
}
|
}
|
||||||
for (k = 0; k < stepSize; k++) { gOutFIFO[k] = gOutputAccum[k];
|
for (k = 0; k < stepSize; k++) {
|
||||||
}
|
gOutFIFO[k] = gOutputAccum[k];
|
||||||
|
}
|
||||||
|
|
||||||
/* shift accumulator */
|
/* shift accumulator */
|
||||||
memmove(gOutputAccum, gOutputAccum+stepSize, fftFrameSize*sizeof(float));
|
memmove(gOutputAccum, gOutputAccum+stepSize, fftFrameSize*sizeof(double));
|
||||||
|
|
||||||
/* move input FIFO */
|
/* move input FIFO */
|
||||||
for (k = 0; k < inFifoLatency; k++) { gInFIFO[k] = gInFIFO[k+stepSize];
|
for (k = 0; k < inFifoLatency; k++) {
|
||||||
}
|
gInFIFO[k] = gInFIFO[k+stepSize];
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
|
void SMBPitchShift::smbFft(double *fftBuffer, int64_t fftFrameSize, int64_t sign)
|
||||||
/*
|
/*
|
||||||
FFT routine, (C)1996 S.M.Bernsee. Sign = -1 is FFT, 1 is iFFT (inverse)
|
FFT routine, (C)1996 S.M.Bernsee. Sign = -1 is FFT, 1 is iFFT (inverse)
|
||||||
Fills fftBuffer[0...2*fftFrameSize-1] with the Fourier transform of the
|
Fills fftBuffer[0...2*fftFrameSize-1] with the Fourier transform of the
|
||||||
|
@ -243,14 +258,16 @@ void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
|
||||||
of the frequencies of interest is in fftBuffer[0...fftFrameSize].
|
of the frequencies of interest is in fftBuffer[0...fftFrameSize].
|
||||||
*/
|
*/
|
||||||
{
|
{
|
||||||
float wr, wi, arg, *p1, *p2, temp;
|
double wr, wi, arg, *p1, *p2, temp;
|
||||||
float tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
|
double tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
|
||||||
long i, bitm, j, le, le2, k;
|
int64_t i, bitm, j, le, le2, k, logN;
|
||||||
|
logN = (int64_t)(log(fftFrameSize) / log(2.) + .5);
|
||||||
|
|
||||||
for (i = 2; i < 2*fftFrameSize-2; i += 2) {
|
for (i = 2; i < 2*fftFrameSize-2; i += 2) {
|
||||||
for (bitm = 2, j = 0; bitm < 2*fftFrameSize; bitm <<= 1) {
|
for (bitm = 2, j = 0; bitm < 2*fftFrameSize; bitm <<= 1) {
|
||||||
if (i & bitm) { j++;
|
if (i & bitm) {
|
||||||
}
|
j++;
|
||||||
|
}
|
||||||
j <<= 1;
|
j <<= 1;
|
||||||
}
|
}
|
||||||
if (i < j) {
|
if (i < j) {
|
||||||
|
@ -260,7 +277,8 @@ void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
|
||||||
*p1 = *p2; *p2 = temp;
|
*p1 = *p2; *p2 = temp;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
for (k = 0, le = 2; k < (long)(log((double)fftFrameSize)/log(2.)+.5); k++) {
|
|
||||||
|
for (k = 0, le = 2; k < logN; k++) {
|
||||||
le <<= 1;
|
le <<= 1;
|
||||||
le2 = le>>1;
|
le2 = le>>1;
|
||||||
ur = 1.0;
|
ur = 1.0;
|
||||||
|
|
|
@ -40,31 +40,33 @@ class SMBPitchShift {
|
||||||
|
|
||||||
float gInFIFO[MAX_FRAME_LENGTH];
|
float gInFIFO[MAX_FRAME_LENGTH];
|
||||||
float gOutFIFO[MAX_FRAME_LENGTH];
|
float gOutFIFO[MAX_FRAME_LENGTH];
|
||||||
float gFFTworksp[2 * MAX_FRAME_LENGTH];
|
double gFFTworksp[2 * MAX_FRAME_LENGTH];
|
||||||
float gLastPhase[MAX_FRAME_LENGTH / 2 + 1];
|
double gLastPhase[MAX_FRAME_LENGTH / 2 + 1];
|
||||||
float gSumPhase[MAX_FRAME_LENGTH / 2 + 1];
|
double gSumPhase[MAX_FRAME_LENGTH / 2 + 1];
|
||||||
float gOutputAccum[2 * MAX_FRAME_LENGTH];
|
double gOutputAccum[2 * MAX_FRAME_LENGTH];
|
||||||
float gAnaFreq[MAX_FRAME_LENGTH];
|
double gAnaFreq[MAX_FRAME_LENGTH];
|
||||||
float gAnaMagn[MAX_FRAME_LENGTH];
|
double gAnaMagn[MAX_FRAME_LENGTH];
|
||||||
float gSynFreq[MAX_FRAME_LENGTH];
|
double gSynFreq[MAX_FRAME_LENGTH];
|
||||||
float gSynMagn[MAX_FRAME_LENGTH];
|
double gSynMagn[MAX_FRAME_LENGTH];
|
||||||
long gRover;
|
int64_t gRover;
|
||||||
|
float lastPitchShift;
|
||||||
|
|
||||||
void smbFft(float *fftBuffer, long fftFrameSize, long sign);
|
void smbFft(double *fftBuffer, int64_t fftFrameSize, int64_t sign);
|
||||||
|
|
||||||
public:
|
public:
|
||||||
void PitchShift(float pitchShift, long numSampsToProcess, long fftFrameSize, long osamp, float sampleRate, float *indata, float *outdata, int stride);
|
void PitchShift(float pitchShift, int64_t numSampsToProcess, int64_t fftFrameSize, int64_t osamp, float sampleRate, float *indata, float *outdata, int stride);
|
||||||
|
|
||||||
SMBPitchShift() {
|
SMBPitchShift() {
|
||||||
gRover = 0;
|
gRover = 0;
|
||||||
memset(gInFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
memset(gInFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
||||||
memset(gOutFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
memset(gOutFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
|
||||||
memset(gFFTworksp, 0, 2 * MAX_FRAME_LENGTH * sizeof(float));
|
memset(gFFTworksp, 0, 2 * MAX_FRAME_LENGTH * sizeof(double));
|
||||||
memset(gLastPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(float));
|
memset(gLastPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(double));
|
||||||
memset(gSumPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(float));
|
memset(gSumPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(double));
|
||||||
memset(gOutputAccum, 0, 2 * MAX_FRAME_LENGTH * sizeof(float));
|
memset(gOutputAccum, 0, 2 * MAX_FRAME_LENGTH * sizeof(double));
|
||||||
memset(gAnaFreq, 0, MAX_FRAME_LENGTH * sizeof(float));
|
memset(gAnaFreq, 0, MAX_FRAME_LENGTH * sizeof(double));
|
||||||
memset(gAnaMagn, 0, MAX_FRAME_LENGTH * sizeof(float));
|
memset(gAnaMagn, 0, MAX_FRAME_LENGTH * sizeof(double));
|
||||||
|
lastPitchShift = 1.0;
|
||||||
}
|
}
|
||||||
};
|
};
|
||||||
|
|
||||||
|
@ -101,9 +103,6 @@ public:
|
||||||
float pitch_scale;
|
float pitch_scale;
|
||||||
int oversampling;
|
int oversampling;
|
||||||
FFT_Size fft_size;
|
FFT_Size fft_size;
|
||||||
float wet;
|
|
||||||
float dry;
|
|
||||||
bool filter;
|
|
||||||
|
|
||||||
protected:
|
protected:
|
||||||
static void _bind_methods();
|
static void _bind_methods();
|
||||||
|
|
Loading…
Reference in New Issue