Merge pull request #57985 from Pineapple/pitchshift-tweaks

PitchShift effect quality and performance tweaks for different pitch scale values

servers/audio/effects/audio_effect_pitch_shift.cpp

@@ -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 freqPerBin, expct;
-	long i,k, qpd, index, inFifoLatency, stepSize, fftFrameSize2;
+	double freqPerBin, expct, reciprocalFftFrameSize;
+	int64_t i,k, qpd, index, inFifoLatency, stepSize, fftFrameSize2;
 
 	/* set up some handy variables */
 	fftFrameSize2 = fftFrameSize/2;
+	reciprocalFftFrameSize = 1./fftFrameSize;
 	stepSize = fftFrameSize/osamp;
-	freqPerBin = sampleRate/(double)fftFrameSize;
-	expct = 2.*Math_PI*(double)stepSize/(double)fftFrameSize;
+	freqPerBin = reciprocalFftFrameSize * sampleRate;
+	expct = Math_TAU * reciprocalFftFrameSize * 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 */
 	for (i = 0; i < numSampsToProcess; i++){
@@ -112,7 +125,7 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 
 			/* do windowing and re,im interleave */
 			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+1] = 0.;
 			}
@@ -124,6 +137,7 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 
 			/* this is the analysis step */
 			for (k = 0; k <= fftFrameSize2; k++) {
+
 				/* de-interlace FFT buffer */
 				real = gFFTworksp[2*k];
 				imag = gFFTworksp[2*k+1];
@@ -141,13 +155,15 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 
 				/* map delta phase into +/- Pi interval */
 				qpd = tmp/Math_PI;
-				if (qpd >= 0) { qpd += qpd&1;
-				} else { qpd -= qpd&1;
-}
+				if (qpd >= 0) {
+					qpd += qpd&1;
+				} else {
+					qpd -= qpd&1;
+				}
 				tmp -= Math_PI*(double)qpd;
 
 				/* 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 */
 				tmp = (double)k*freqPerBin + tmp*freqPerBin;
@@ -160,8 +176,8 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 
 			/* ***************** PROCESSING ******************* */
 			/* this does the actual pitch shifting */
-			memset(gSynMagn, 0, fftFrameSize*sizeof(float));
-			memset(gSynFreq, 0, fftFrameSize*sizeof(float));
+			memset(gSynMagn, 0, fftFrameSize*sizeof(double));
+			memset(gSynFreq, 0, fftFrameSize*sizeof(double));
 			for (k = 0; k <= fftFrameSize2; k++) {
 				index = k*pitchShift;
 				if (index <= fftFrameSize2) {
@@ -184,7 +200,7 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 				tmp /= freqPerBin;
 
 				/* take osamp into account */
-				tmp = 2.*Math_PI*tmp/osamp;
+				tmp = Math_TAU*tmp/osamp;
 
 				/* add the overlap phase advance back in */
 				tmp += (double)k*expct;
@@ -199,33 +215,35 @@ void SMBPitchShift::PitchShift(float pitchShift, long numSampsToProcess, long ff
 			}
 
 			/* 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 */
 			smbFft(gFFTworksp, fftFrameSize, 1);
 
 			/* do windowing and add to output accumulator */
 			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);
 			}
-			for (k = 0; k < stepSize; k++) { gOutFIFO[k] = gOutputAccum[k];
-}
+			for (k = 0; k < stepSize; k++) {
+				gOutFIFO[k] = gOutputAccum[k];
+			}
 
 			/* shift accumulator */
-			memmove(gOutputAccum, gOutputAccum+stepSize, fftFrameSize*sizeof(float));
+			memmove(gOutputAccum, gOutputAccum+stepSize, fftFrameSize*sizeof(double));
 
 			/* 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)
 	Fills fftBuffer[0...2*fftFrameSize-1] with the Fourier transform of the
@@ -238,14 +256,16 @@ void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
 	of the frequencies of interest is in fftBuffer[0...fftFrameSize].
 */
 {
-	float wr, wi, arg, *p1, *p2, temp;
-	float tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
-	long i, bitm, j, le, le2, k;
+	double wr, wi, arg, *p1, *p2, temp;
+	double tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
+	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 (bitm = 2, j = 0; bitm < 2*fftFrameSize; bitm <<= 1) {
-			if (i & bitm) { j++;
-}
+			if (i & bitm) {
+				j++;
+			}
 			j <<= 1;
 		}
 		if (i < j) {
@@ -255,7 +275,8 @@ void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
 			*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;
 		le2 = le>>1;
 		ur = 1.0;
@@ -288,6 +309,14 @@ void SMBPitchShift::smbFft(float *fftBuffer, long fftFrameSize, long sign)
 void AudioEffectPitchShiftInstance::process(const AudioFrame *p_src_frames, AudioFrame *p_dst_frames, int p_frame_count) {
 	float sample_rate = AudioServer::get_singleton()->get_mix_rate();
 
+	// For pitch_scale 1.0 it's cheaper to just pass samples without processing them.
+	if (Math::is_equal_approx(base->pitch_scale, 1.0f)) {
+		for (int i = 0; i < p_frame_count; i++) {
+			p_dst_frames[i] = p_src_frames[i];
+		}
+		return;
+	}
+
 	float *in_l = (float *)p_src_frames;
 	float *in_r = in_l + 1;
 
@@ -361,7 +390,4 @@ AudioEffectPitchShift::AudioEffectPitchShift() {
 	pitch_scale = 1.0;
 	oversampling = 4;
 	fft_size = FFT_SIZE_2048;
-	wet = 0.0;
-	dry = 0.0;
-	filter = false;
 }

servers/audio/effects/audio_effect_pitch_shift.h

@@ -40,31 +40,33 @@ class SMBPitchShift {
 
 	float gInFIFO[MAX_FRAME_LENGTH];
 	float gOutFIFO[MAX_FRAME_LENGTH];
-	float gFFTworksp[2 * MAX_FRAME_LENGTH];
-	float gLastPhase[MAX_FRAME_LENGTH / 2 + 1];
-	float gSumPhase[MAX_FRAME_LENGTH / 2 + 1];
-	float gOutputAccum[2 * MAX_FRAME_LENGTH];
-	float gAnaFreq[MAX_FRAME_LENGTH];
-	float gAnaMagn[MAX_FRAME_LENGTH];
-	float gSynFreq[MAX_FRAME_LENGTH];
-	float gSynMagn[MAX_FRAME_LENGTH];
-	long gRover;
-
-	void smbFft(float *fftBuffer, long fftFrameSize, long sign);
+	double gFFTworksp[2 * MAX_FRAME_LENGTH];
+	double gLastPhase[MAX_FRAME_LENGTH / 2 + 1];
+	double gSumPhase[MAX_FRAME_LENGTH / 2 + 1];
+	double gOutputAccum[2 * MAX_FRAME_LENGTH];
+	double gAnaFreq[MAX_FRAME_LENGTH];
+	double gAnaMagn[MAX_FRAME_LENGTH];
+	double gSynFreq[MAX_FRAME_LENGTH];
+	double gSynMagn[MAX_FRAME_LENGTH];
+	int64_t gRover;
+	float lastPitchShift;
+
+	void smbFft(double *fftBuffer, int64_t fftFrameSize, int64_t sign);
 
 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() {
 		gRover = 0;
 		memset(gInFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
 		memset(gOutFIFO, 0, MAX_FRAME_LENGTH * sizeof(float));
-		memset(gFFTworksp, 0, 2 * MAX_FRAME_LENGTH * sizeof(float));
-		memset(gLastPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(float));
-		memset(gSumPhase, 0, (MAX_FRAME_LENGTH / 2 + 1) * sizeof(float));
-		memset(gOutputAccum, 0, 2 * MAX_FRAME_LENGTH * sizeof(float));
-		memset(gAnaFreq, 0, MAX_FRAME_LENGTH * sizeof(float));
-		memset(gAnaMagn, 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));
+		lastPitchShift = 1.0;
 	}
 };
 
@@ -101,9 +103,6 @@ public:
 	float pitch_scale;
 	int oversampling;
 	FFTSize fft_size;
-	float wet;
-	float dry;
-	bool filter;
 
 protected:
 	static void _bind_methods();