Update SoundTouch to 2.3.2 commit 1eda9c0b01039f29d230a46cda9f2290bbd1f62b

Externals/soundtouch/AAFilter.cpp

@@ -12,13 +12,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-05 23:40:22 +0200 (Sun, 05 Jan 2014) $
-// File revision : $Revision: 4 $
-//
-// $Id: AAFilter.cpp 177 2014-01-05 21:40:22Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -49,7 +42,7 @@
 
 using namespace soundtouch;
 
-#define PI        3.141592655357989
+#define PI       3.14159265358979323846
 #define TWOPI    (2 * PI)
 
 // define this to save AA filter coefficients to a file
@@ -61,7 +54,7 @@ using namespace soundtouch;
     static void _DEBUG_SAVE_AAFIR_COEFFS(SAMPLETYPE *coeffs, int len)
     {
         FILE *fptr = fopen("aa_filter_coeffs.txt", "wt");
-        if (fptr == NULL) return;
+        if (fptr == nullptr) return;
 
         for (int i = 0; i < len; i ++)
         {
@@ -75,7 +68,6 @@ using namespace soundtouch;
     #define _DEBUG_SAVE_AAFIR_COEFFS(x, y)
 #endif
 
-
 /*****************************************************************************
  *
  * Implementation of the class 'AAFilter'
@@ -90,14 +82,12 @@ AAFilter::AAFilter(uint len)
 }
 
 
-
 AAFilter::~AAFilter()
 {
     delete pFIR;
 }
 
 
-
 // Sets new anti-alias filter cut-off edge frequency, scaled to
 // sampling frequency (nyquist frequency = 0.5).
 // The filter will cut frequencies higher than the given frequency.
@@ -108,7 +98,6 @@ void AAFilter::setCutoffFreq(double newCutoffFreq)
 }
 
 
-
 // Sets number of FIR filter taps
 void AAFilter::setLength(uint newLength)
 {
@@ -117,7 +106,6 @@ void AAFilter::setLength(uint newLength)
 }
 
 
-
 // Calculates coefficients for a low-pass FIR filter using Hamming window
 void AAFilter::calculateCoeffs()
 {
@@ -177,12 +165,10 @@ void AAFilter::calculateCoeffs()
     for (i = 0; i < length; i ++) 
     {
         temp = work[i] * scaleCoeff;
-//#if SOUNDTOUCH_INTEGER_SAMPLES
         // scale & round to nearest integer
         temp += (temp >= 0) ? 0.5 : -0.5;
         // ensure no overfloods
         assert(temp >= -32768 && temp <= 32767);
-//#endif
         coeffs[i] = (SAMPLETYPE)temp;
     }
 

Externals/soundtouch/AAFilter.h

@@ -13,13 +13,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-07 21:41:23 +0200 (Tue, 07 Jan 2014) $
-// File revision : $Revision: 4 $
-//
-// $Id: AAFilter.h 187 2014-01-07 19:41:23Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library

Externals/soundtouch/BPMDetect.cpp

@@ -26,13 +26,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: BPMDetect.cpp 202 2015-02-21 21:24:29Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -54,45 +47,62 @@
 //
 ////////////////////////////////////////////////////////////////////////////////
 
+#define _USE_MATH_DEFINES
+
 #include <math.h>
 #include <assert.h>
 #include <string.h>
 #include <stdio.h>
+#include <cfloat>
 #include "FIFOSampleBuffer.h"
 #include "PeakFinder.h"
 #include "BPMDetect.h"
 
 using namespace soundtouch;
 
-#define INPUT_BLOCK_SAMPLES       2048
+// algorithm input sample block size
-#define DECIMATED_BLOCK_SAMPLES   256
+static const int INPUT_BLOCK_SIZE = 2048;
 
-/// decay constant for calculating RMS volume sliding average approximation 
+// decimated sample block size
-/// (time constant is about 10 sec)
+static const int DECIMATED_BLOCK_SIZE = 256;
-const float avgdecay = 0.99986f;
 
-/// Normalization coefficient for calculating RMS sliding average approximation.
+/// Target sample rate after decimation
-const float avgnorm = (1 - avgdecay);
+static const int TARGET_SRATE = 1000;
 
+/// XCorr update sequence size, update in about 200msec chunks
+static const int XCORR_UPDATE_SEQUENCE = (int)(TARGET_SRATE / 5);
+
+/// Moving average N size
+static const int MOVING_AVERAGE_N = 15;
+
+/// XCorr decay time constant, decay to half in 30 seconds
+/// If it's desired to have the system adapt quicker to beat rate 
+/// changes within a continuing music stream, then the 
+/// 'xcorr_decay_time_constant' value can be reduced, yet that
+/// can increase possibility of glitches in bpm detection.
+static const double XCORR_DECAY_TIME_CONSTANT = 30.0;
+
+/// Data overlap factor for beat detection algorithm
+static const int OVERLAP_FACTOR = 4;
+
+static const double TWOPI = (2 * M_PI);
 
 ////////////////////////////////////////////////////////////////////////////////
 
 // Enable following define to create bpm analysis file:
 
-// #define _CREATE_BPM_DEBUG_FILE
+//#define _CREATE_BPM_DEBUG_FILE
 
 #ifdef _CREATE_BPM_DEBUG_FILE
 
-    #define DEBUGFILE_NAME  "c:\\temp\\soundtouch-bpm-debug.txt"
+    static void _SaveDebugData(const char *name, const float *data, int minpos, int maxpos, double coeff)
-
-    static void _SaveDebugData(const float *data, int minpos, int maxpos, double coeff)
     {
-        FILE *fptr = fopen(DEBUGFILE_NAME, "wt");
+        FILE *fptr = fopen(name, "wt");
         int i;
 
         if (fptr)
         {
-            printf("\n\nWriting BPM debug data into file " DEBUGFILE_NAME "\n\n");
+            printf("\nWriting BPM debug data into file %s\n", name);
             for (i = minpos; i < maxpos; i ++)
             {
                 fprintf(fptr, "%d\t%.1lf\t%f\n", i, coeff / (double)i, data[i]);
@@ -100,42 +110,90 @@ const float avgnorm = (1 - avgdecay);
             fclose(fptr);
         }
     }
+
+    void _SaveDebugBeatPos(const char *name, const std::vector<BEAT> &beats)
+    {
+        printf("\nWriting beat detections data into file %s\n", name);
+
+        FILE *fptr = fopen(name, "wt");
+        if (fptr)
+        {
+            for (uint i = 0; i < beats.size(); i++)
+            {
+                BEAT b = beats[i];
+                fprintf(fptr, "%lf\t%lf\n", b.pos, b.strength);
+            }
+            fclose(fptr);
+        }
+    }
 #else
-    #define _SaveDebugData(a,b,c,d)
+    #define _SaveDebugData(name, a,b,c,d)
+    #define _SaveDebugBeatPos(name, b)
 #endif
 
+// Hamming window
+void hamming(float *w, int N)
+{
+    for (int i = 0; i < N; i++)
+    {
+        w[i] = (float)(0.54 - 0.46 * cos(TWOPI * i / (N - 1)));
+    }
+
+}
+
+////////////////////////////////////////////////////////////////////////////////
+//
+// IIR2_filter - 2nd order IIR filter
+
+IIR2_filter::IIR2_filter(const double *lpf_coeffs)
+{
+    memcpy(coeffs, lpf_coeffs, 5 * sizeof(double));
+    memset(prev, 0, sizeof(prev));
+}
+
+
+float IIR2_filter::update(float x)
+{
+    prev[0] = x;
+    double y = x * coeffs[0];
+
+    for (int i = 4; i >= 1; i--)
+    {
+        y += coeffs[i] * prev[i];
+        prev[i] = prev[i - 1];
+    }
+
+    prev[3] = y;
+    return (float)y;
+}
+
+
+// IIR low-pass filter coefficients, calculated with matlab/octave cheby2(2,40,0.05)
+const double _LPF_coeffs[5] = { 0.00996655391939, -0.01944529148401, 0.00996655391939, 1.96867605796247, -0.96916387431724 };
+
 ////////////////////////////////////////////////////////////////////////////////
 
-
+BPMDetect::BPMDetect(int numChannels, int aSampleRate) :
-BPMDetect::BPMDetect(int numChannels, int aSampleRate)
+    beat_lpf(_LPF_coeffs)
 {
+    beats.reserve(250); // initial reservation to prevent frequent reallocation
+
     this->sampleRate = aSampleRate;
     this->channels = numChannels;
 
     decimateSum = 0;
     decimateCount = 0;
 
-    envelopeAccu = 0;
-
-    // Initialize RMS volume accumulator to RMS level of 1500 (out of 32768) that's
-    // safe initial RMS signal level value for song data. This value is then adapted
-    // to the actual level during processing.
-#ifdef SOUNDTOUCH_INTEGER_SAMPLES
-    // integer samples
-    RMSVolumeAccu = (1500 * 1500) / avgnorm;
-#else
-    // float samples, scaled to range [-1..+1[
-    RMSVolumeAccu = (0.045f * 0.045f) / avgnorm;
-#endif
-
     // choose decimation factor so that result is approx. 1000 Hz
-    decimateBy = sampleRate / 1000;
+    decimateBy = sampleRate / TARGET_SRATE;
-    assert(decimateBy > 0);
+    if ((decimateBy <= 0) || (decimateBy * DECIMATED_BLOCK_SIZE < INPUT_BLOCK_SIZE))
-    assert(INPUT_BLOCK_SAMPLES < decimateBy * DECIMATED_BLOCK_SAMPLES);
+    {
+        ST_THROW_RT_ERROR("Too small samplerate");
+    }
 
     // Calculate window length & starting item according to desired min & max bpms
     windowLen = (60 * sampleRate) / (decimateBy * MIN_BPM);
-    windowStart = (60 * sampleRate) / (decimateBy * MAX_BPM);
+    windowStart = (60 * sampleRate) / (decimateBy * MAX_BPM_RANGE);
 
     assert(windowLen > windowStart);
 
@@ -143,23 +201,38 @@ BPMDetect::BPMDetect(int numChannels, int aSampleRate)
     xcorr = new float[windowLen];
     memset(xcorr, 0, windowLen * sizeof(float));
 
+    pos = 0;
+    peakPos = 0;
+    peakVal = 0;
+    init_scaler = 1;
+    beatcorr_ringbuffpos = 0;
+    beatcorr_ringbuff = new float[windowLen];
+    memset(beatcorr_ringbuff, 0, windowLen * sizeof(float));
+
     // allocate processing buffer
     buffer = new FIFOSampleBuffer();
     // we do processing in mono mode
     buffer->setChannels(1);
     buffer->clear();
-}
 
+    // calculate hamming windows
+    hamw = new float[XCORR_UPDATE_SEQUENCE];
+    hamming(hamw, XCORR_UPDATE_SEQUENCE);
+    hamw2 = new float[XCORR_UPDATE_SEQUENCE / 2];
+    hamming(hamw2, XCORR_UPDATE_SEQUENCE / 2);
+}
 
 
 BPMDetect::~BPMDetect()
 {
     delete[] xcorr;
+    delete[] beatcorr_ringbuff;
+    delete[] hamw;
+    delete[] hamw2;
     delete buffer;
 }
 
 
-
 /// convert to mono, low-pass filter & decimate to about 500 Hz. 
 /// return number of outputted samples.
 ///
@@ -216,7 +289,6 @@ int BPMDetect::decimate(SAMPLETYPE *dest, const SAMPLETYPE *src, int numsamples)
 }
 
 
-
 // Calculates autocorrelation function of the sample history buffer
 void BPMDetect::updateXCorr(int process_samples)
 {
@@ -224,72 +296,122 @@ void BPMDetect::updateXCorr(int process_samples)
     SAMPLETYPE *pBuffer;
     
     assert(buffer->numSamples() >= (uint)(process_samples + windowLen));
+    assert(process_samples == XCORR_UPDATE_SEQUENCE);
 
     pBuffer = buffer->ptrBegin();
+
+    // calculate decay factor for xcorr filtering
+    float xcorr_decay = (float)pow(0.5, 1.0 / (XCORR_DECAY_TIME_CONSTANT * TARGET_SRATE / process_samples));
+
+    // prescale pbuffer
+    float tmp[XCORR_UPDATE_SEQUENCE];
+    for (int i = 0; i < process_samples; i++)
+    {
+        tmp[i] = hamw[i] * hamw[i] * pBuffer[i];
+    }
+
     #pragma omp parallel for
     for (offs = windowStart; offs < windowLen; offs ++) 
     {
-        LONG_SAMPLETYPE sum;
+        float sum;
         int i;
 
         sum = 0;
         for (i = 0; i < process_samples; i ++) 
         {
-            sum += pBuffer[i] * pBuffer[i + offs];    // scaling the sub-result shouldn't be necessary
+            sum += tmp[i] * pBuffer[i + offs];  // scaling the sub-result shouldn't be necessary
         }
-//        xcorr[offs] *= xcorr_decay;   // decay 'xcorr' here with suitable coefficients 
+        xcorr[offs] *= xcorr_decay;   // decay 'xcorr' here with suitable time constant.
-                                        // if it's desired that the system adapts automatically to
-                                        // various bpms, e.g. in processing continouos music stream.
-                                        // The 'xcorr_decay' should be a value that's smaller than but 
-                                        // close to one, and should also depend on 'process_samples' value.
 
-        xcorr[offs] += (float)sum;
+        xcorr[offs] += (float)fabs(sum);
     }
 }
 
 
-// Calculates envelope of the sample data
+// Detect individual beat positions
-void BPMDetect::calcEnvelope(SAMPLETYPE *samples, int numsamples) 
+void BPMDetect::updateBeatPos(int process_samples)
 {
-    const static double decay = 0.7f;               // decay constant for smoothing the envelope
+    SAMPLETYPE *pBuffer;
-    const static double norm = (1 - decay);
 
-    int i;
+    assert(buffer->numSamples() >= (uint)(process_samples + windowLen));
-    LONG_SAMPLETYPE out;
-    double val;
 
-    for (i = 0; i < numsamples; i ++) 
+    pBuffer = buffer->ptrBegin();
+    assert(process_samples == XCORR_UPDATE_SEQUENCE / 2);
+
+    //    static double thr = 0.0003;
+    double posScale = (double)this->decimateBy / (double)this->sampleRate;
+    int resetDur = (int)(0.12 / posScale + 0.5);
+
+    // prescale pbuffer
+    float tmp[XCORR_UPDATE_SEQUENCE / 2];
+    for (int i = 0; i < process_samples; i++)
     {
-        // calc average RMS volume
+        tmp[i] = hamw2[i] * hamw2[i] * pBuffer[i];
-        RMSVolumeAccu *= avgdecay;
-        val = (float)fabs((float)samples[i]);
-        RMSVolumeAccu += val * val;
-
-        // cut amplitudes that are below cutoff ~2 times RMS volume
-        // (we're interested in peak values, not the silent moments)
-        if (val < 0.5 * sqrt(RMSVolumeAccu * avgnorm))
-        {
-            val = 0;
     }
 
-        // smooth amplitude envelope
+    #pragma omp parallel for
-        envelopeAccu *= decay;
+    for (int offs = windowStart; offs < windowLen; offs++)
-        envelopeAccu += val;
+    {
-        out = (LONG_SAMPLETYPE)(envelopeAccu * norm);
+        float sum = 0;
+        for (int i = 0; i < process_samples; i++)
+        {
+            sum += tmp[i] * pBuffer[offs + i];
+        }
+        beatcorr_ringbuff[(beatcorr_ringbuffpos + offs) % windowLen] += (float)((sum > 0) ? sum : 0); // accumulate only positive correlations
+    }
 
-#ifdef SOUNDTOUCH_INTEGER_SAMPLES
+    int skipstep = XCORR_UPDATE_SEQUENCE / OVERLAP_FACTOR;
-        // cut peaks (shouldn't be necessary though)
+
-        if (out > 32767) out = 32767;
+    // compensate empty buffer at beginning by scaling coefficient
-#endif // SOUNDTOUCH_INTEGER_SAMPLES
+    float scale = (float)windowLen / (float)(skipstep * init_scaler);
-        samples[i] = (SAMPLETYPE)out;
+    if (scale > 1.0f)
+    {
+        init_scaler++;
+    }
+    else
+    {
+        scale = 1.0f;
+    }
+
+    // detect beats
+    for (int i = 0; i < skipstep; i++)
+    {
+        float sum = beatcorr_ringbuff[beatcorr_ringbuffpos];
+        sum -= beat_lpf.update(sum);
+
+        if (sum > peakVal)
+        {
+            // found new local largest value
+            peakVal = sum;
+            peakPos = pos;
+        }
+        if (pos > peakPos + resetDur)
+        {
+            // largest value not updated for 200msec => accept as beat
+            peakPos += skipstep;
+            if (peakVal > 0)
+            {
+                // add detected beat to end of "beats" vector
+                BEAT temp = { (float)(peakPos * posScale), (float)(peakVal * scale) };
+                beats.push_back(temp);
+            }
+
+            peakVal = 0;
+            peakPos = pos;
+        }
+
+        beatcorr_ringbuff[beatcorr_ringbuffpos] = 0;
+        pos++;
+        beatcorr_ringbuffpos = (beatcorr_ringbuffpos + 1) % windowLen;
     }
 }
 
 
+#define max(x,y) ((x) > (y) ? (x) : (y))
 
 void BPMDetect::inputSamples(const SAMPLETYPE *samples, int numSamples)
 {
-    SAMPLETYPE decimated[DECIMATED_BLOCK_SAMPLES];
+    SAMPLETYPE decimated[DECIMATED_BLOCK_SIZE];
 
     // iterate so that max INPUT_BLOCK_SAMPLES processed per iteration
     while (numSamples > 0)
@@ -297,48 +419,70 @@ void BPMDetect::inputSamples(const SAMPLETYPE *samples, int numSamples)
         int block;
         int decSamples;
 
-        block = (numSamples > INPUT_BLOCK_SAMPLES) ? INPUT_BLOCK_SAMPLES : numSamples;
+        block = (numSamples > INPUT_BLOCK_SIZE) ? INPUT_BLOCK_SIZE : numSamples;
 
         // decimate. note that converts to mono at the same time
         decSamples = decimate(decimated, samples, block);
         samples += block * channels;
         numSamples -= block;
 
-        // envelope new samples and add them to buffer
-        calcEnvelope(decimated, decSamples);
         buffer->putSamples(decimated, decSamples);
     }
 
-    // when the buffer has enought samples for processing...
+    // when the buffer has enough samples for processing...
-    if ((int)buffer->numSamples() > windowLen) 
+    int req = max(windowLen + XCORR_UPDATE_SEQUENCE, 2 * XCORR_UPDATE_SEQUENCE);
+    while ((int)buffer->numSamples() >= req) 
     {
-        int processLength;
+        // ... update autocorrelations...
-
+        updateXCorr(XCORR_UPDATE_SEQUENCE);
-        // how many samples are processed
+        // ...update beat position calculation...
-        processLength = (int)buffer->numSamples() - windowLen;
+        updateBeatPos(XCORR_UPDATE_SEQUENCE / 2);
-
+        // ... and remove proceessed samples from the buffer
-        // ... calculate autocorrelations for oldest samples...
+        int n = XCORR_UPDATE_SEQUENCE / OVERLAP_FACTOR;
-        updateXCorr(processLength);
+        buffer->receiveSamples(n);
-        // ... and remove them from the buffer
-        buffer->receiveSamples(processLength);
     }
 }
 
 
-
 void BPMDetect::removeBias()
 {
     int i;
-    float minval = 1e12f;   // arbitrary large number
 
+    // Remove linear bias: calculate linear regression coefficient
+    // 1. calc mean of 'xcorr' and 'i'
+    double mean_i = 0;
+    double mean_x = 0;
+    for (i = windowStart; i < windowLen; i++)
+    {
+        mean_x += xcorr[i];
+    }
+    mean_x /= (windowLen - windowStart);
+    mean_i = 0.5 * (windowLen - 1 + windowStart);
+
+    // 2. calculate linear regression coefficient
+    double b = 0;
+    double div = 0;
+    for (i = windowStart; i < windowLen; i++)
+    {
+        double xt = xcorr[i] - mean_x;
+        double xi = i - mean_i;
+        b += xt * xi;
+        div += xi * xi;
+    }
+    b /= div;
+
+    // subtract linear regression and resolve min. value bias
+    float minval = FLT_MAX;   // arbitrary large number
     for (i = windowStart; i < windowLen; i ++)
     {
+        xcorr[i] -= (float)(b * i);
         if (xcorr[i] < minval)
         {
             minval = xcorr[i];
         }
     }
 
+    // subtract min.value
     for (i = windowStart; i < windowLen; i ++)
     {
         xcorr[i] -= minval;
@@ -346,26 +490,82 @@ void BPMDetect::removeBias()
 }
 
 
+// Calculate N-point moving average for "source" values
+void MAFilter(float *dest, const float *source, int start, int end, int N)
+{
+    for (int i = start; i < end; i++)
+    {
+        int i1 = i - N / 2;
+        int i2 = i + N / 2 + 1;
+        if (i1 < start) i1 = start;
+        if (i2 > end)   i2 = end;
+
+        double sum = 0;
+        for (int j = i1; j < i2; j ++)
+        { 
+            sum += source[j];
+        }
+        dest[i] = (float)(sum / (i2 - i1));
+    }
+}
+
+
 float BPMDetect::getBpm()
 {
     double peakPos;
     double coeff;
     PeakFinder peakFinder;
 
-    coeff = 60.0 * ((double)sampleRate / (double)decimateBy);
-
-    // save bpm debug analysis data if debug data enabled
-    _SaveDebugData(xcorr, windowStart, windowLen, coeff);
-
     // remove bias from xcorr data
     removeBias();
 
+    coeff = 60.0 * ((double)sampleRate / (double)decimateBy);
+
+    // save bpm debug data if debug data writing enabled
+    _SaveDebugData("soundtouch-bpm-xcorr.txt", xcorr, windowStart, windowLen, coeff);
+
+    // Smoothen by N-point moving-average
+    float *data = new float[windowLen];
+    memset(data, 0, sizeof(float) * windowLen);
+    MAFilter(data, xcorr, windowStart, windowLen, MOVING_AVERAGE_N);
+
     // find peak position
-    peakPos = peakFinder.detectPeak(xcorr, windowStart, windowLen);
+    peakPos = peakFinder.detectPeak(data, windowStart, windowLen);
+
+    // save bpm debug data if debug data writing enabled
+    _SaveDebugData("soundtouch-bpm-smoothed.txt", data, windowStart, windowLen, coeff);
+
+    delete[] data;
 
     assert(decimateBy != 0);
     if (peakPos < 1e-9) return 0.0; // detection failed.
 
+    _SaveDebugBeatPos("soundtouch-detected-beats.txt", beats);
+
     // calculate BPM
-    return (float) (coeff / peakPos);
+    float bpm = (float)(coeff / peakPos);
+    return (bpm >= MIN_BPM && bpm <= MAX_BPM_VALID) ? bpm : 0;
+}
+
+
+/// Get beat position arrays. Note: The array includes also really low beat detection values 
+/// in absence of clear strong beats. Consumer may wish to filter low values away.
+/// - "pos" receive array of beat positions
+/// - "values" receive array of beat detection strengths
+/// - max_num indicates max.size of "pos" and "values" array.  
+///
+/// You can query a suitable array sized by calling this with nullptr in "pos" & "values".
+///
+/// \return number of beats in the arrays.
+int BPMDetect::getBeats(float *pos, float *values, int max_num)
+{
+    int num = (int)beats.size();
+    if ((!pos) || (!values)) return num;    // pos or values nullptr, return just size
+
+    for (int i = 0; (i < num) && (i < max_num); i++)
+    {
+        pos[i] = beats[i].pos;
+        values[i] = beats[i].strength;
+    }
+    return num;
 }

Externals/soundtouch/BPMDetect.h

@@ -26,13 +26,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-08-30 22:53:44 +0300 (Thu, 30 Aug 2012) $
-// File revision : $Revision: 4 $
-//
-// $Id: BPMDetect.h 150 2012-08-30 19:53:44Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -57,32 +50,49 @@
 #ifndef _BPMDetect_H_
 #define _BPMDetect_H_
 
+#include <vector>
 #include "STTypes.h"
 #include "FIFOSampleBuffer.h"
 
 namespace soundtouch
 {
 
-/// Minimum allowed BPM rate. Used to restrict accepted result above a reasonable limit.
+    /// Minimum allowed BPM rate. Used to restrict accepted result above a reasonable limit.
-#define MIN_BPM 29
+    #define MIN_BPM 45
 
-/// Maximum allowed BPM rate. Used to restrict accepted result below a reasonable limit.
+    /// Maximum allowed BPM rate range. Used for calculating algorithm parametrs
-#define MAX_BPM 200
+    #define MAX_BPM_RANGE 200
+
+    /// Maximum allowed BPM rate range. Used to restrict accepted result below a reasonable limit.
+    #define MAX_BPM_VALID 190
+
+////////////////////////////////////////////////////////////////////////////////
+
+    typedef struct
+    {
+        float pos;
+        float strength;
+    } BEAT;
 
 
-/// Class for calculating BPM rate for audio data.
+    class IIR2_filter
-class BPMDetect
+    {
-{
+        double coeffs[5];
-protected:
+        double prev[5];
+
+    public:
+        IIR2_filter(const double *lpf_coeffs);
+        float update(float x);
+    };
+
+
+    /// Class for calculating BPM rate for audio data.
+    class BPMDetect
+    {
+    protected:
         /// Auto-correlation accumulator bins.
         float *xcorr;
 
-    /// Amplitude envelope sliding average approximation level accumulator
-    double envelopeAccu;
-
-    /// RMS volume sliding average approximation level accumulator
-    double RMSVolumeAccu;
-
         /// Sample average counter.
         int decimateCount;
 
@@ -105,9 +115,28 @@ protected:
         /// the first these many correlation bins.
         int windowStart;
 
+        /// window functions for data preconditioning
+        float *hamw;
+        float *hamw2;
+
+        // beat detection variables
+        int pos;
+        int peakPos;
+        int beatcorr_ringbuffpos;
+        int init_scaler;
+        float peakVal;
+        float *beatcorr_ringbuff;
+
         /// FIFO-buffer for decimated processing samples.
         soundtouch::FIFOSampleBuffer *buffer;
 
+        /// Collection of detected beat positions
+        //BeatCollection beats;
+        std::vector<BEAT> beats;
+
+        // 2nd order low-pass-filter
+        IIR2_filter beat_lpf;
+
         /// Updates auto-correlation function for given number of decimated samples that 
         /// are read from the internal 'buffer' pipe (samples aren't removed from the pipe 
         /// though).
@@ -131,7 +160,11 @@ protected:
         /// remove constant bias from xcorr data
         void removeBias();
 
-public:
+        // Detect individual beat positions
+        void updateBeatPos(int process_samples);
+
+
+    public:
         /// Constructor.
         BPMDetect(int numChannels,  ///< Number of channels in sample data.
             int sampleRate    ///< Sample rate in Hz.
@@ -150,15 +183,23 @@ public:
             int numSamples                            ///< Number of samples in buffer
         );
 
-
         /// Analyzes the results and returns the BPM rate. Use this function to read result
         /// after whole song data has been input to the class by consecutive calls of
         /// 'inputSamples' function.
         ///
         /// \return Beats-per-minute rate, or zero if detection failed.
         float getBpm();
-};
 
+        /// Get beat position arrays. Note: The array includes also really low beat detection values 
+        /// in absence of clear strong beats. Consumer may wish to filter low values away.
+        /// - "pos" receive array of beat positions
+        /// - "values" receive array of beat detection strengths
+        /// - max_num indicates max.size of "pos" and "values" array.  
+        ///
+        /// You can query a suitable array sized by calling this with nullptr in "pos" & "values".
+        ///
+        /// \return number of beats in the arrays.
+        int getBeats(float *pos, float *strength, int max_num);
+    };
 }
-
 #endif // _BPMDetect_H_

Externals/soundtouch/CMakeLists.txt

@@ -1,3 +1,6 @@
+# OSX meeds to know
+check_and_add_flag(CXX11 -std=c++11)
+
 set(SRCS
 	AAFilter.cpp
 	BPMDetect.cpp
@@ -17,4 +20,3 @@ set(SRCS
 
 add_library(SoundTouch STATIC ${SRCS})
 dolphin_disable_warnings_msvc(SoundTouch)
-add_definitions(-w)

Externals/soundtouch/FIFOSampleBuffer.cpp

@@ -15,13 +15,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-11-08 20:53:01 +0200 (Thu, 08 Nov 2012) $
-// File revision : $Revision: 4 $
-//
-// $Id: FIFOSampleBuffer.cpp 160 2012-11-08 18:53:01Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -57,8 +50,8 @@ FIFOSampleBuffer::FIFOSampleBuffer(int numChannels)
 {
     assert(numChannels > 0);
     sizeInBytes = 0; // reasonable initial value
-    buffer = NULL;
+    buffer = nullptr;
-    bufferUnaligned = NULL;
+    bufferUnaligned = nullptr;
     samplesInBuffer = 0;
     bufferPos = 0;
     channels = (uint)numChannels;
@@ -70,8 +63,8 @@ FIFOSampleBuffer::FIFOSampleBuffer(int numChannels)
 FIFOSampleBuffer::~FIFOSampleBuffer()
 {
     delete[] bufferUnaligned;
-    bufferUnaligned = NULL;
+    bufferUnaligned = nullptr;
-    buffer = NULL;
+    buffer = nullptr;
 }
 
 
@@ -80,7 +73,8 @@ void FIFOSampleBuffer::setChannels(int numChannels)
 {
     uint usedBytes;
 
-    assert(numChannels > 0);
+    if (!verifyNumberOfChannels(numChannels)) return;
+
     usedBytes = channels * samplesInBuffer;
     channels = (uint)numChannels;
     samplesInBuffer = usedBytes / channels;
@@ -131,7 +125,7 @@ void FIFOSampleBuffer::putSamples(uint nSamples)
 //
 // Parameter 'slackCapacity' tells the function how much free capacity (in
 // terms of samples) there _at least_ should be, in order to the caller to
-// succesfully insert all the required samples to the buffer. When necessary, 
+// successfully insert all the required samples to the buffer. When necessary, 
 // the function grows the buffer size to comply with this requirement.
 //
 // When using this function as means for inserting new samples, also remember 
@@ -158,7 +152,7 @@ SAMPLETYPE *FIFOSampleBuffer::ptrBegin()
 }
 
 
-// Ensures that the buffer has enought capacity, i.e. space for _at least_
+// Ensures that the buffer has enough capacity, i.e. space for _at least_
 // 'capacityRequirement' number of samples. The buffer is grown in steps of
 // 4 kilobytes to eliminate the need for frequently growing up the buffer,
 // as well as to round the buffer size up to the virtual memory page size.
@@ -172,7 +166,7 @@ void FIFOSampleBuffer::ensureCapacity(uint capacityRequirement)
         sizeInBytes = (capacityRequirement * channels * sizeof(SAMPLETYPE) + 4095) & (uint)-4096;
         assert(sizeInBytes % 2 == 0);
         tempUnaligned = new SAMPLETYPE[sizeInBytes / sizeof(SAMPLETYPE) + 16 / sizeof(SAMPLETYPE)];
-        if (tempUnaligned == NULL)
+        if (tempUnaligned == nullptr)
         {
             ST_THROW_RT_ERROR("Couldn't allocate memory!\n");
         }
@@ -272,3 +266,10 @@ uint FIFOSampleBuffer::adjustAmountOfSamples(uint numSamples)
     return samplesInBuffer;
 }
 
+
+/// Add silence to end of buffer
+void FIFOSampleBuffer::addSilent(uint nSamples)
+{
+    memset(ptrEnd(nSamples), 0, sizeof(SAMPLETYPE) * nSamples * channels);
+    samplesInBuffer += nSamples;
+}

Externals/soundtouch/FIFOSampleBuffer.h

@@ -15,13 +15,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-05 23:40:22 +0200 (Sun, 05 Jan 2014) $
-// File revision : $Revision: 4 $
-//
-// $Id: FIFOSampleBuffer.h 177 2014-01-05 21:40:22Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -98,7 +91,7 @@ public:
                      );
 
     /// destructor
-    ~FIFOSampleBuffer();
+    ~FIFOSampleBuffer() override;
 
     /// Returns a pointer to the beginning of the output samples. 
     /// This function is provided for accessing the output samples directly. 
@@ -107,7 +100,7 @@ public:
     /// When using this function to output samples, also remember to 'remove' the
     /// output samples from the buffer by calling the 
     /// 'receiveSamples(numSamples)' function
-    virtual SAMPLETYPE *ptrBegin();
+    virtual SAMPLETYPE *ptrBegin() override;
 
     /// Returns a pointer to the end of the used part of the sample buffer (i.e. 
     /// where the new samples are to be inserted). This function may be used for 
@@ -119,7 +112,7 @@ public:
     /// 'putSamples(numSamples)' function.
     SAMPLETYPE *ptrEnd(
                 uint slackCapacity   ///< How much free capacity (in samples) there _at least_ 
-                                     ///< should be so that the caller can succesfully insert the 
+                                     ///< should be so that the caller can successfully insert the 
                                      ///< desired samples to the buffer. If necessary, the function 
                                      ///< grows the buffer size to comply with this requirement.
                 );
@@ -128,7 +121,7 @@ public:
     /// the sample buffer.
     virtual void putSamples(const SAMPLETYPE *samples,  ///< Pointer to samples.
                             uint numSamples                         ///< Number of samples to insert.
-                            );
+                            ) override;
 
     /// Adjusts the book-keeping to increase number of samples in the buffer without 
     /// copying any actual samples.
@@ -146,7 +139,7 @@ public:
     /// \return Number of samples returned.
     virtual uint receiveSamples(SAMPLETYPE *output, ///< Buffer where to copy output samples.
                                 uint maxSamples                 ///< How many samples to receive at max.
-                                );
+                                ) override;
 
     /// Adjusts book-keeping so that given number of samples are removed from beginning of the 
     /// sample buffer without copying them anywhere. 
@@ -154,10 +147,10 @@ public:
     /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly
     /// with 'ptrBegin' function.
     virtual uint receiveSamples(uint maxSamples   ///< Remove this many samples from the beginning of pipe.
-                                );
+                                ) override;
 
     /// Returns number of samples currently available.
-    virtual uint numSamples() const;
+    virtual uint numSamples() const override;
 
     /// Sets number of channels, 1 = mono, 2 = stereo.
     void setChannels(int numChannels);
@@ -169,14 +162,17 @@ public:
     }
 
     /// Returns nonzero if there aren't any samples available for outputting.
-    virtual int isEmpty() const;
+    virtual int isEmpty() const override;
 
     /// Clears all the samples.
-    virtual void clear();
+    virtual void clear() override;
 
     /// allow trimming (downwards) amount of samples in pipeline.
     /// Returns adjusted amount of samples
-    uint adjustAmountOfSamples(uint numSamples);
+    uint adjustAmountOfSamples(uint numSamples) override;
+
+    /// Add silence to end of buffer
+    void addSilent(uint nSamples);
 };
 
 }

Externals/soundtouch/FIFOSamplePipe.h

@@ -17,13 +17,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-06-13 22:29:53 +0300 (Wed, 13 Jun 2012) $
-// File revision : $Revision: 4 $
-//
-// $Id: FIFOSamplePipe.h 143 2012-06-13 19:29:53Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -58,6 +51,18 @@ namespace soundtouch
 /// Abstract base class for FIFO (first-in-first-out) sample processing classes.
 class FIFOSamplePipe
 {
+protected:
+
+    bool verifyNumberOfChannels(int nChannels) const
+    {
+        if ((nChannels > 0) && (nChannels <= SOUNDTOUCH_MAX_CHANNELS))
+        {
+            return true;
+        }
+        ST_THROW_RT_ERROR("Error: Illegal number of channels");
+        return false;
+    }
+
 public:
     // virtual default destructor
     virtual ~FIFOSamplePipe() {}
@@ -122,7 +127,6 @@ public:
 };
 
 
-
 /// Base-class for sound processing routines working in FIFO principle. With this base 
 /// class it's easy to implement sound processing stages that can be chained together,
 /// so that samples that are fed into beginning of the pipe automatically go through 
@@ -140,20 +144,18 @@ protected:
     /// Sets output pipe.
     void setOutPipe(FIFOSamplePipe *pOutput)
     {
-        assert(output == NULL);
+        assert(output == nullptr);
-        assert(pOutput != NULL);
+        assert(pOutput != nullptr);
         output = pOutput;
     }
 
-
     /// Constructor. Doesn't define output pipe; it has to be set be 
     /// 'setOutPipe' function.
     FIFOProcessor()
     {
-        output = NULL;
+        output = nullptr;
     }
 
-
     /// Constructor. Configures output pipe.
     FIFOProcessor(FIFOSamplePipe *pOutput   ///< Output pipe.
                  )
@@ -161,13 +163,11 @@ protected:
         output = pOutput;
     }
 
-
     /// Destructor.
-    virtual ~FIFOProcessor()
+    virtual ~FIFOProcessor() override
     {
     }
 
-
     /// Returns a pointer to the beginning of the output samples. 
     /// This function is provided for accessing the output samples directly. 
     /// Please be careful for not to corrupt the book-keeping!
@@ -175,7 +175,7 @@ protected:
     /// When using this function to output samples, also remember to 'remove' the
     /// output samples from the buffer by calling the 
     /// 'receiveSamples(numSamples)' function
-    virtual SAMPLETYPE *ptrBegin()
+    virtual SAMPLETYPE *ptrBegin() override
     {
         return output->ptrBegin();
     }
@@ -189,44 +189,40 @@ public:
     /// \return Number of samples returned.
     virtual uint receiveSamples(SAMPLETYPE *outBuffer, ///< Buffer where to copy output samples.
                                 uint maxSamples                    ///< How many samples to receive at max.
-                                )
+                                ) override
     {
         return output->receiveSamples(outBuffer, maxSamples);
     }
 
-
     /// Adjusts book-keeping so that given number of samples are removed from beginning of the 
     /// sample buffer without copying them anywhere. 
     ///
     /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly
     /// with 'ptrBegin' function.
     virtual uint receiveSamples(uint maxSamples   ///< Remove this many samples from the beginning of pipe.
-                                )
+                                ) override
     {
         return output->receiveSamples(maxSamples);
     }
 
-
     /// Returns number of samples currently available.
-    virtual uint numSamples() const
+    virtual uint numSamples() const override
     {
         return output->numSamples();
     }
 
-
     /// Returns nonzero if there aren't any samples available for outputting.
-    virtual int isEmpty() const
+    virtual int isEmpty() const override
     {
         return output->isEmpty();
     }
 
     /// allow trimming (downwards) amount of samples in pipeline.
     /// Returns adjusted amount of samples
-    virtual uint adjustAmountOfSamples(uint numSamples)
+    virtual uint adjustAmountOfSamples(uint numSamples) override
     {
         return output->adjustAmountOfSamples(numSamples);
     }
-
 };
 
 }

Externals/soundtouch/FIRFilter.cpp

@@ -2,22 +2,21 @@
 ///
 /// General FIR digital filter routines with MMX optimization. 
 ///
-/// Note : MMX optimized functions reside in a separate, platform-specific file, 
+/// Notes : MMX optimized functions reside in a separate, platform-specific file, 
 /// e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp'
 ///
+/// This source file contains OpenMP optimizations that allow speeding up the
+/// corss-correlation algorithm by executing it in several threads / CPU cores 
+/// in parallel. See the following article link for more detailed discussion 
+/// about SoundTouch OpenMP optimizations:
+/// http://www.softwarecoven.com/parallel-computing-in-embedded-mobile-devices
+///
 /// Author        : Copyright (c) Olli Parviainen
 /// Author e-mail : oparviai 'at' iki.fi
 /// SoundTouch WWW: http://www.surina.net/soundtouch
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: FIRFilter.cpp 202 2015-02-21 21:24:29Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -60,53 +59,43 @@ FIRFilter::FIRFilter()
     resultDivider = 0;
     length = 0;
     lengthDiv8 = 0;
-    filterCoeffs = NULL;
+    filterCoeffs = nullptr;
+    filterCoeffsStereo = nullptr;
 }
 
 
 FIRFilter::~FIRFilter()
 {
     delete[] filterCoeffs;
+    delete[] filterCoeffsStereo;
 }
 
+
 // Usual C-version of the filter routine for stereo sound
 uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
 {
     int j, end;
-#ifdef SOUNDTOUCH_FLOAT_SAMPLES
+    // hint compiler autovectorization that loop length is divisible by 8
-    // when using floating point samples, use a scaler instead of a divider
+    uint ilength = length & -8;
-    // because division is much slower operation than multiplying.
-    double dScaler = 1.0 / (double)resultDivider;
-#endif
 
-    assert(length != 0);
+    assert((length != 0) && (length == ilength) && (src != nullptr) && (dest != nullptr) && (filterCoeffs != nullptr));
-    assert(src != NULL);
+    assert(numSamples > ilength);
-    assert(dest != NULL);
-    assert(filterCoeffs != NULL);
 
-    end = 2 * (numSamples - length);
+    end = 2 * (numSamples - ilength);
 
     #pragma omp parallel for
     for (j = 0; j < end; j += 2) 
     {
         const SAMPLETYPE *ptr;
         LONG_SAMPLETYPE suml, sumr;
-        uint i;
 
         suml = sumr = 0;
         ptr = src + j;
 
-        for (i = 0; i < length; i += 4) 
+        for (uint i = 0; i < ilength; i ++)
         {
-            // loop is unrolled by factor of 4 here for efficiency
+            suml += ptr[2 * i] * filterCoeffsStereo[2 * i];
-            suml += ptr[2 * i + 0] * filterCoeffs[i + 0] +
+            sumr += ptr[2 * i + 1] * filterCoeffsStereo[2 * i + 1];
-                    ptr[2 * i + 2] * filterCoeffs[i + 1] +
-                    ptr[2 * i + 4] * filterCoeffs[i + 2] +
-                    ptr[2 * i + 6] * filterCoeffs[i + 3];
-            sumr += ptr[2 * i + 1] * filterCoeffs[i + 0] +
-                    ptr[2 * i + 3] * filterCoeffs[i + 1] +
-                    ptr[2 * i + 5] * filterCoeffs[i + 2] +
-                    ptr[2 * i + 7] * filterCoeffs[i + 3];
         }
 
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
@@ -116,54 +105,41 @@ uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, ui
         suml = (suml < -32768) ? -32768 : (suml > 32767) ? 32767 : suml;
         // saturate to 16 bit integer limits
         sumr = (sumr < -32768) ? -32768 : (sumr > 32767) ? 32767 : sumr;
-#else
-        suml *= dScaler;
-        sumr *= dScaler;
 #endif // SOUNDTOUCH_INTEGER_SAMPLES
         dest[j] = (SAMPLETYPE)suml;
         dest[j + 1] = (SAMPLETYPE)sumr;
     }
-    return numSamples - length;
+    return numSamples - ilength;
 }
 
 
-
-
 // Usual C-version of the filter routine for mono sound
 uint FIRFilter::evaluateFilterMono(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
 {
     int j, end;
-#ifdef SOUNDTOUCH_FLOAT_SAMPLES
-    // when using floating point samples, use a scaler instead of a divider
-    // because division is much slower operation than multiplying.
-    double dScaler = 1.0 / (double)resultDivider;
-#endif
 
-    assert(length != 0);
+    // hint compiler autovectorization that loop length is divisible by 8
+    int ilength = length & -8;
 
-    end = numSamples - length;
+    assert(ilength != 0);
+
+    end = numSamples - ilength;
     #pragma omp parallel for
     for (j = 0; j < end; j ++)
     {
         const SAMPLETYPE *pSrc = src + j;
         LONG_SAMPLETYPE sum;
-        uint i;
+        int i;
 
         sum = 0;
-        for (i = 0; i < length; i += 4) 
+        for (i = 0; i < ilength; i ++)
         {
-            // loop is unrolled by factor of 4 here for efficiency
+            sum += pSrc[i] * filterCoeffs[i];
-            sum += pSrc[i + 0] * filterCoeffs[i + 0] + 
-                   pSrc[i + 1] * filterCoeffs[i + 1] + 
-                   pSrc[i + 2] * filterCoeffs[i + 2] + 
-                   pSrc[i + 3] * filterCoeffs[i + 3];
         }
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
         sum >>= resultDivFactor;
         // saturate to 16 bit integer limits
         sum = (sum < -32768) ? -32768 : (sum > 32767) ? 32767 : sum;
-#else
-        sum *= dScaler;
 #endif // SOUNDTOUCH_INTEGER_SAMPLES
         dest[j] = (SAMPLETYPE)sum;
     }
@@ -175,26 +151,24 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
 {
     int j, end;
 
-#ifdef SOUNDTOUCH_FLOAT_SAMPLES
-    // when using floating point samples, use a scaler instead of a divider
-    // because division is much slower operation than multiplying.
-    double dScaler = 1.0 / (double)resultDivider;
-#endif
-
     assert(length != 0);
-    assert(src != NULL);
+    assert(src != nullptr);
-    assert(dest != NULL);
+    assert(dest != nullptr);
-    assert(filterCoeffs != NULL);
+    assert(filterCoeffs != nullptr);
     assert(numChannels < 16);
 
-    end = numChannels * (numSamples - length);
+    // hint compiler autovectorization that loop length is divisible by 8
+    int ilength = length & -8;
+
+    end = numChannels * (numSamples - ilength);
 
     #pragma omp parallel for
     for (j = 0; j < end; j += numChannels)
     {
         const SAMPLETYPE *ptr;
         LONG_SAMPLETYPE sums[16];
-        uint c, i;
+        uint c;
+        int i;
 
         for (c = 0; c < numChannels; c ++)
         {
@@ -203,7 +177,7 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
 
         ptr = src + j;
 
-        for (i = 0; i < length; i ++)
+        for (i = 0; i < ilength; i ++)
         {
             SAMPLETYPE coef=filterCoeffs[i];
             for (c = 0; c < numChannels; c ++)
@@ -217,13 +191,11 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
         {
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
             sums[c] >>= resultDivFactor;
-#else
-            sums[c] *= dScaler;
 #endif // SOUNDTOUCH_INTEGER_SAMPLES
             dest[j+c] = (SAMPLETYPE)sums[c];
         }
     }
-    return numSamples - length;
+    return numSamples - ilength;
 }
 
 
@@ -235,6 +207,13 @@ void FIRFilter::setCoefficients(const SAMPLETYPE *coeffs, uint newLength, uint u
     assert(newLength > 0);
     if (newLength % 8) ST_THROW_RT_ERROR("FIR filter length not divisible by 8");
 
+    #ifdef SOUNDTOUCH_FLOAT_SAMPLES
+        // scale coefficients already here if using floating samples
+        double scale = 1.0 / resultDivider;
+    #else
+        short scale = 1;
+    #endif
+
     lengthDiv8 = newLength / 8;
     length = lengthDiv8 * 8;
     assert(length == newLength);
@@ -244,7 +223,16 @@ void FIRFilter::setCoefficients(const SAMPLETYPE *coeffs, uint newLength, uint u
 
     delete[] filterCoeffs;
     filterCoeffs = new SAMPLETYPE[length];
-    memcpy(filterCoeffs, coeffs, length * sizeof(SAMPLETYPE));
+    delete[] filterCoeffsStereo;
+    filterCoeffsStereo = new SAMPLETYPE[length*2];
+    for (uint i = 0; i < length; i ++)
+    {
+        filterCoeffs[i] = (SAMPLETYPE)(coeffs[i] * scale);
+        // create also stereo set of filter coefficients: this allows compiler
+        // to autovectorize filter evaluation much more efficiently
+        filterCoeffsStereo[2 * i] = (SAMPLETYPE)(coeffs[i] * scale);
+        filterCoeffsStereo[2 * i + 1] = (SAMPLETYPE)(coeffs[i] * scale);
+    }
 }
 
 
@@ -254,7 +242,6 @@ uint FIRFilter::getLength() const
 }
 
 
-
 // Applies the filter to the given sequence of samples. 
 //
 // Note : The amount of outputted samples is by value of 'filter_length' 
@@ -284,10 +271,9 @@ uint FIRFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSample
 }
 
 
-
 // Operator 'new' is overloaded so that it automatically creates a suitable instance 
 // depending on if we've a MMX-capable CPU available or not.
-void * FIRFilter::operator new(size_t s)
+void * FIRFilter::operator new(size_t)
 {
     // Notice! don't use "new FIRFilter" directly, use "newInstance" to create a new instance instead!
     ST_THROW_RT_ERROR("Error in FIRFilter::new: Don't use 'new FIRFilter', use 'newInstance' member instead!");

Externals/soundtouch/FIRFilter.h

@@ -11,13 +11,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: FIRFilter.h 202 2015-02-21 21:24:29Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -64,6 +57,7 @@ protected:
 
     // Memory for filter coefficients
     SAMPLETYPE *filterCoeffs;
+    SAMPLETYPE *filterCoeffsStereo;
 
     virtual uint evaluateFilterStereo(SAMPLETYPE *dest, 
                                       const SAMPLETYPE *src, 
@@ -112,12 +106,12 @@ public:
         short *filterCoeffsUnalign;
         short *filterCoeffsAlign;
 
-        virtual uint evaluateFilterStereo(short *dest, const short *src, uint numSamples) const;
+        virtual uint evaluateFilterStereo(short *dest, const short *src, uint numSamples) const override;
     public:
         FIRFilterMMX();
         ~FIRFilterMMX();
 
-        virtual void setCoefficients(const short *coeffs, uint newLength, uint uResultDivFactor);
+        virtual void setCoefficients(const short *coeffs, uint newLength, uint uResultDivFactor) override;
     };
 
 #endif // SOUNDTOUCH_ALLOW_MMX
@@ -131,12 +125,12 @@ public:
         float *filterCoeffsUnalign;
         float *filterCoeffsAlign;
 
-        virtual uint evaluateFilterStereo(float *dest, const float *src, uint numSamples) const;
+        virtual uint evaluateFilterStereo(float *dest, const float *src, uint numSamples) const override;
     public:
         FIRFilterSSE();
         ~FIRFilterSSE();
 
-        virtual void setCoefficients(const float *coeffs, uint newLength, uint uResultDivFactor);
+        virtual void setCoefficients(const float *coeffs, uint newLength, uint uResultDivFactor) override;
     };
 
 #endif // SOUNDTOUCH_ALLOW_SSE

Externals/soundtouch/InterpolateCubic.cpp

@@ -8,10 +8,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateCubic.cpp 179 2014-01-06 18:41:42Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library

Externals/soundtouch/InterpolateCubic.h

@@ -8,10 +8,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateCubic.h 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -45,21 +41,27 @@ namespace soundtouch
 class InterpolateCubic : public TransposerBase
 {
 protected:
-    virtual void resetRegisters();
     virtual int transposeMono(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
     virtual int transposeStereo(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
     virtual int transposeMulti(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
 
     double fract;
 
 public:
     InterpolateCubic();
+
+    virtual void resetRegisters() override;
+
+    virtual int getLatency() const override
+    {
+        return 1;
+    }
 };
 
 }

Externals/soundtouch/InterpolateLinear.cpp

@@ -8,10 +8,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateLinear.cpp 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -146,7 +142,7 @@ int InterpolateLinearInteger::transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE
         LONG_SAMPLETYPE temp, vol1;
     
         assert(iFract < SCALE);
-        vol1 = (SCALE - iFract);
+        vol1 = (LONG_SAMPLETYPE)(SCALE - iFract);
         for (int c = 0; c < numChannels; c ++)
         {
             temp = vol1 * src[c] + iFract * src[c + numChannels];

Externals/soundtouch/InterpolateLinear.h

@@ -8,10 +8,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateLinear.h 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -42,39 +38,42 @@
 namespace soundtouch
 {
 
-/// Linear transposer class that uses integer arithmetics
+/// Linear transposer class that uses integer arithmetic
 class InterpolateLinearInteger : public TransposerBase
 {
 protected:
     int iFract;
     int iRate;
 
-    virtual void resetRegisters();
-
     virtual int transposeMono(SAMPLETYPE *dest, 
                        const SAMPLETYPE *src, 
-                       int &srcSamples);
+                       int &srcSamples) override;
     virtual int transposeStereo(SAMPLETYPE *dest, 
                          const SAMPLETYPE *src, 
-                         int &srcSamples);
+                         int &srcSamples) override;
-    virtual int transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples);
+    virtual int transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, int &srcSamples) override;
 public:
     InterpolateLinearInteger();
 
     /// Sets new target rate. Normal rate = 1.0, smaller values represent slower 
     /// rate, larger faster rates.
-    virtual void setRate(double newRate);
+    virtual void setRate(double newRate) override;
+
+    virtual void resetRegisters() override;
+
+    virtual int getLatency() const override
+    {
+        return 0;
+    }
 };
 
 
-/// Linear transposer class that uses floating point arithmetics
+/// Linear transposer class that uses floating point arithmetic
 class InterpolateLinearFloat : public TransposerBase
 {
 protected:
     double fract;
 
-    virtual void resetRegisters();
-
     virtual int transposeMono(SAMPLETYPE *dest, 
                        const SAMPLETYPE *src, 
                        int &srcSamples);
@@ -85,6 +84,13 @@ protected:
 
 public:
     InterpolateLinearFloat();
+
+    virtual void resetRegisters();
+
+    int getLatency() const
+    {
+        return 0;
+    }
 };
 
 }

Externals/soundtouch/InterpolateShannon.cpp

@@ -13,10 +13,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateShannon.cpp 195 2014-04-06 15:57:21Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -175,9 +171,9 @@ int InterpolateShannon::transposeStereo(SAMPLETYPE *pdest,
 
 /// Transpose stereo audio. Returns number of produced output samples, and 
 /// updates "srcSamples" to amount of consumed source samples
-int InterpolateShannon::transposeMulti(SAMPLETYPE *pdest, 
+int InterpolateShannon::transposeMulti(SAMPLETYPE *, 
-                    const SAMPLETYPE *psrc, 
+                    const SAMPLETYPE *, 
-                    int &srcSamples)
+                    int &)
 {
     // not implemented
     assert(false);

Externals/soundtouch/InterpolateShannon.h

@@ -13,10 +13,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateShannon.h 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -50,21 +46,27 @@ namespace soundtouch
 class InterpolateShannon : public TransposerBase
 {
 protected:
-    void resetRegisters();
     int transposeMono(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
     int transposeStereo(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
     int transposeMulti(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
-                        int &srcSamples);
+                        int &srcSamples) override;
 
     double fract;
 
 public:
     InterpolateShannon();
+
+    void resetRegisters() override;
+
+    virtual int getLatency() const override
+    {
+        return 3;
+    }
 };
 
 }

Externals/soundtouch/PeakFinder.cpp

@@ -11,13 +11,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-05-18 18:22:02 +0300 (Mon, 18 May 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: PeakFinder.cpp 213 2015-05-18 15:22:02Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -149,7 +142,7 @@ int PeakFinder::findCrossingLevel(const float *data, float level, int peakpos, i
     peaklevel = data[peakpos];
     assert(peaklevel >= level);
     pos = peakpos;
-    while ((pos >= minPos) && (pos < maxPos))
+    while ((pos >= minPos) && (pos + direction < maxPos))
     {
         if (data[pos + direction] < level) return pos;   // crossing found
         pos += direction;
@@ -178,7 +171,6 @@ double PeakFinder::calcMassCenter(const float *data, int firstPos, int lastPos)
 }
 
 
-
 /// get exact center of peak near given position by calculating local mass of center
 double PeakFinder::getPeakCenter(const float *data, int peakpos) const
 {
@@ -218,7 +210,6 @@ double PeakFinder::getPeakCenter(const float *data, int peakpos) const
 }
 
 
-
 double PeakFinder::detectPeak(const float *data, int aminPos, int amaxPos) 
 {
 
@@ -249,12 +240,12 @@ double PeakFinder::detectPeak(const float *data, int aminPos, int amaxPos)
     // - sometimes the highest peak can be Nth harmonic of the true base peak yet 
     // just a slightly higher than the true base
 
-    for (i = 3; i < 10; i ++)
+    for (i = 1; i < 3; i ++)
     {
         double peaktmp, harmonic;
         int i1,i2;
 
-        harmonic = (double)i * 0.5;
+        harmonic = (double)pow(2.0, i);
         peakpos = (int)(highPeak / harmonic + 0.5f);
         if (peakpos < minPos) break;
         peakpos = findTop(data, peakpos);   // seek true local maximum index

Externals/soundtouch/PeakFinder.h

@@ -9,13 +9,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2011-12-30 22:33:46 +0200 (Fri, 30 Dec 2011) $
-// File revision : $Revision: 4 $
-//
-// $Id: PeakFinder.h 132 2011-12-30 20:33:46Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -51,8 +44,8 @@ protected:
 
     /// Calculates the mass center between given vector items.
     double calcMassCenter(const float *data, ///< Data vector.
-                         int firstPos,      ///< Index of first vector item beloging to the peak.
+                         int firstPos,      ///< Index of first vector item belonging to the peak.
-                         int lastPos        ///< Index of last vector item beloging to the peak.
+                         int lastPos        ///< Index of last vector item belonging to the peak.
                          ) const;
 
     /// Finds the data vector index where the monotoniously decreasing signal crosses the

Externals/soundtouch/RateTransposer.cpp

@@ -10,13 +10,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: RateTransposer.cpp 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -57,15 +50,21 @@ TransposerBase::ALGORITHM TransposerBase::algorithm = TransposerBase::CUBIC;
 // Constructor
 RateTransposer::RateTransposer() : FIFOProcessor(&outputBuffer)
 {
-    bUseAAFilter = true;
+    bUseAAFilter = 
+#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
+        true;
+#else
+        // Disable Anti-alias filter if desirable to avoid click at rate change zero value crossover
+        false;
+#endif
 
     // Instantiates the anti-alias filter
     pAAFilter = new AAFilter(64);
     pTransposer = TransposerBase::newInstance();
+    clear();
 }
 
 
-
 RateTransposer::~RateTransposer()
 {
     delete pAAFilter;
@@ -73,11 +72,14 @@ RateTransposer::~RateTransposer()
 }
 
 
-
 /// Enables/disables the anti-alias filter. Zero to disable, nonzero to enable
 void RateTransposer::enableAAFilter(bool newMode)
 {
+#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
+    // Disable Anti-alias filter if desirable to avoid click at rate change zero value crossover
     bUseAAFilter = newMode;
+    clear();
+#endif
 }
 
 
@@ -94,7 +96,6 @@ AAFilter *RateTransposer::getAAFilter()
 }
 
 
-
 // Sets new target iRate. Normal iRate = 1.0, smaller values represent slower 
 // iRate, larger faster iRates.
 void RateTransposer::setRate(double newRate)
@@ -130,8 +131,6 @@ void RateTransposer::putSamples(const SAMPLETYPE *samples, uint nSamples)
 // the 'set_returnBuffer_size' function.
 void RateTransposer::processSamples(const SAMPLETYPE *src, uint nSamples)
 {
-    uint count;
-
     if (nSamples == 0) return;
 
     // Store samples to input buffer
@@ -141,7 +140,7 @@ void RateTransposer::processSamples(const SAMPLETYPE *src, uint nSamples)
     // the filter
     if (bUseAAFilter == false) 
     {
-        count = pTransposer->transpose(outputBuffer, inputBuffer);
+        (void)pTransposer->transpose(outputBuffer, inputBuffer);
         return;
     }
 
@@ -177,11 +176,10 @@ void RateTransposer::processSamples(const SAMPLETYPE *src, uint nSamples)
 // Sets the number of channels, 1 = mono, 2 = stereo
 void RateTransposer::setChannels(int nChannels)
 {
-    assert(nChannels > 0);
+    if (!verifyNumberOfChannels(nChannels) ||
+        (pTransposer->numChannels == nChannels)) return;
 
-    if (pTransposer->numChannels == nChannels) return;
     pTransposer->setChannels(nChannels);
-
     inputBuffer.setChannels(nChannels);
     midBuffer.setChannels(nChannels);
     outputBuffer.setChannels(nChannels);
@@ -194,6 +192,11 @@ void RateTransposer::clear()
     outputBuffer.clear();
     midBuffer.clear();
     inputBuffer.clear();
+    pTransposer->resetRegisters();
+
+    // prefill buffer to avoid losing first samples at beginning of stream
+    int prefill = getLatency();
+    inputBuffer.addSilent(prefill);
 }
 
 
@@ -208,6 +211,14 @@ int RateTransposer::isEmpty() const
 }
 
 
+/// Return approximate initial input-output latency
+int RateTransposer::getLatency() const
+{
+    return pTransposer->getLatency() +
+        ((bUseAAFilter) ? (pAAFilter->getLength() / 2) : 0);
+}
+
+
 //////////////////////////////////////////////////////////////////////////////
 //
 // TransposerBase - Base class for interpolation
@@ -280,7 +291,7 @@ void TransposerBase::setRate(double newRate)
 TransposerBase *TransposerBase::newInstance()
 {
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
-    // Notice: For integer arithmetics support only linear algorithm (due to simplest calculus)
+    // Notice: For integer arithmetic support only linear algorithm (due to simplest calculus)
     return ::new InterpolateLinearInteger;
 #else
     switch (algorithm)
@@ -296,7 +307,7 @@ TransposerBase *TransposerBase::newInstance()
 
         default:
             assert(false);
-            return NULL;
+            return nullptr;
     }
 #endif
 }

Externals/soundtouch/RateTransposer.h

@@ -14,13 +14,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: RateTransposer.h 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -66,8 +59,6 @@ public:
     };
 
 protected:
-    virtual void resetRegisters() = 0;
-
     virtual int transposeMono(SAMPLETYPE *dest, 
                         const SAMPLETYPE *src, 
                         int &srcSamples)  = 0;
@@ -90,6 +81,9 @@ public:
     virtual int transpose(FIFOSampleBuffer &dest, FIFOSampleBuffer &src);
     virtual void setRate(double newRate);
     virtual void setChannels(int channels);
+    virtual int getLatency() const = 0;
+
+    virtual void resetRegisters() = 0;
 
     // static factory function
     static TransposerBase *newInstance();
@@ -130,23 +124,11 @@ protected:
 
 public:
     RateTransposer();
-    virtual ~RateTransposer();
+    virtual ~RateTransposer() override;
-
-    /// Operator 'new' is overloaded so that it automatically creates a suitable instance 
-    /// depending on if we're to use integer or floating point arithmetics.
-//    static void *operator new(size_t s);
-
-    /// Use this function instead of "new" operator to create a new instance of this class. 
-    /// This function automatically chooses a correct implementation, depending on if 
-    /// integer ot floating point arithmetics are to be used.
-//    static RateTransposer *newInstance();
 
     /// Returns the output buffer object
     FIFOSamplePipe *getOutput() { return &outputBuffer; };
 
-    /// Returns the store buffer object
-//    FIFOSamplePipe *getStore() { return &storeBuffer; };
-
     /// Return anti-alias filter object
     AAFilter *getAAFilter();
 
@@ -165,13 +147,16 @@ public:
 
     /// Adds 'numSamples' pcs of samples from the 'samples' memory position into
     /// the input of the object.
-    void putSamples(const SAMPLETYPE *samples, uint numSamples);
+    void putSamples(const SAMPLETYPE *samples, uint numSamples) override;
 
     /// Clears all the samples in the object
-    void clear();
+    void clear() override;
 
     /// Returns nonzero if there aren't any samples available for outputting.
-    int isEmpty() const;
+    int isEmpty() const override;
+
+    /// Return approximate initial input-output latency
+    int getLatency() const;
 };
 
 }

Externals/soundtouch/STTypes.h

@@ -8,13 +8,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-05-18 18:25:07 +0300 (Mon, 18 May 2015) $
-// File revision : $Revision: 3 $
-//
-// $Id: STTypes.h 215 2015-05-18 15:25:07Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -39,33 +32,37 @@
 #ifndef STTypes_H
 #define STTypes_H
 
+typedef unsigned int    uint;
+typedef unsigned long   ulong;
+
+// Patch for MinGW: on Win64 long is 32-bit
+#ifdef _WIN64
+    typedef unsigned long long ulongptr;
+#else
+    typedef ulong ulongptr;
+#endif
+
+
+// Helper macro for aligning pointer up to next 16-byte boundary
+#define SOUNDTOUCH_ALIGN_POINTER_16(x)      ( ( (ulongptr)(x) + 15 ) & ~(ulongptr)15 )
+
+
 #if (defined(__GNUC__) && !defined(ANDROID))
     // In GCC, include soundtouch_config.h made by config scritps.
     // Skip this in Android compilation that uses GCC but without configure scripts.
-    //#include "soundtouch_config.h"
+    #include "soundtouch_config.h"
 #endif
 
 
 namespace soundtouch
 {
-    typedef unsigned int    uint;
+    /// Max allowed number of channels
-    typedef unsigned long   ulong;
+    #define SOUNDTOUCH_MAX_CHANNELS     16
-
-    // Patch for MinGW: on Win64 long is 32-bit
-    #ifdef _WIN64
-        typedef unsigned long long ulongptr;
-    #else
-        typedef ulong ulongptr;
-    #endif
-
-
-    // Helper macro for aligning pointer up to next 16-byte boundary
-    #define SOUNDTOUCH_ALIGN_POINTER_16(x)      ( ( (ulongptr)(x) + 15 ) & ~(ulongptr)15 )
 
     /// Activate these undef's to overrule the possible sampletype 
     /// setting inherited from some other header file:
-    #undef SOUNDTOUCH_INTEGER_SAMPLES
+    //#undef SOUNDTOUCH_INTEGER_SAMPLES
-    #undef SOUNDTOUCH_FLOAT_SAMPLES
+    //#undef SOUNDTOUCH_FLOAT_SAMPLES
 
     /// If following flag is defined, always uses multichannel processing 
     /// routines also for mono and stero sound. This is for routine testing 
@@ -74,7 +71,7 @@ namespace soundtouch
     /// runtime performance so recommendation is to keep this off.
     // #define USE_MULTICH_ALWAYS
 
-    #if (defined(__SOFTFP__))
+    #if (defined(__SOFTFP__) && defined(ANDROID))
         // For Android compilation: Force use of Integer samples in case that
         // compilation uses soft-floating point emulation - soft-fp is way too slow
         #undef  SOUNDTOUCH_FLOAT_SAMPLES
@@ -97,8 +94,8 @@ namespace soundtouch
         ///   However, if you still prefer to select the sample format here 
         ///   also in GNU environment, then please #undef the INTEGER_SAMPLE
         ///   and FLOAT_SAMPLE defines first as in comments above.
-        #define SOUNDTOUCH_INTEGER_SAMPLES     1    //< 16bit integer samples
+        //#define SOUNDTOUCH_INTEGER_SAMPLES     1    //< 16bit integer samples
-        //#define SOUNDTOUCH_FLOAT_SAMPLES       1    //< 32bit float samples
+        #define SOUNDTOUCH_FLOAT_SAMPLES       1    //< 32bit float samples
      
     #endif
 
@@ -109,7 +106,7 @@ namespace soundtouch
         /// routines compiled for whatever reason, you may disable these optimizations 
         /// to make the library compile.
 
-        //#define SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS     1
+        #define SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS     1
 
         /// In GNU environment, allow the user to override this setting by
         /// giving the following switch to the configure script:
@@ -124,10 +121,10 @@ namespace soundtouch
 
     #endif
 
-    // If defined, allows the SIMD-optimized routines to take minor shortcuts 
+    // If defined, allows the SIMD-optimized routines to skip unevenly aligned
-    // for improved performance. Undefine to require faithfully similar SIMD 
+    // memory offsets that can cause performance penalty in some SIMD implementations.
-    // calculations as in normal C implementation.
+    // Causes slight compromise in sound quality.
-    #define SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION    1
+    // #define SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION    1
 
 
     #ifdef SOUNDTOUCH_INTEGER_SAMPLES
@@ -142,16 +139,19 @@ namespace soundtouch
         #endif // SOUNDTOUCH_FLOAT_SAMPLES
 
         #ifdef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS
-            // Allow MMX optimizations
+            // Allow MMX optimizations (not available in X64 mode)
+            #if (!_M_X64)
                 #define SOUNDTOUCH_ALLOW_MMX   1
             #endif
+        #endif
 
     #else
 
         // floating point samples
         typedef float  SAMPLETYPE;
-        // data type for sample accumulation: Use double to utilize full precision.
+        // data type for sample accumulation: Use float also here to enable
-        typedef double LONG_SAMPLETYPE;
+        // efficient autovectorization
+        typedef float LONG_SAMPLETYPE;
 
         #ifdef SOUNDTOUCH_ALLOW_X86_OPTIMIZATIONS
             // Allow SSE optimizations
@@ -160,10 +160,16 @@ namespace soundtouch
 
     #endif  // SOUNDTOUCH_INTEGER_SAMPLES
 
-};
+    #if ((SOUNDTOUCH_ALLOW_SSE) || (__SSE__) || (SOUNDTOUCH_USE_NEON))
+        #if SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION
+            #define ST_SIMD_AVOID_UNALIGNED
+        #endif
+    #endif
+
+}
 
 // define ST_NO_EXCEPTION_HANDLING switch to disable throwing std exceptions:
-   #define ST_NO_EXCEPTION_HANDLING    1
+// #define ST_NO_EXCEPTION_HANDLING    1
 #ifdef ST_NO_EXCEPTION_HANDLING
     // Exceptions disabled. Throw asserts instead if enabled.
     #include <assert.h>

Externals/soundtouch/SoundTouch.cpp

@@ -41,13 +41,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: SoundTouch.cpp 225 2015-07-26 14:45:48Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -118,7 +111,6 @@ SoundTouch::SoundTouch()
 }
 
 
-
 SoundTouch::~SoundTouch()
 {
     delete pRateTransposer;
@@ -126,7 +118,6 @@ SoundTouch::~SoundTouch()
 }
 
 
-
 /// Get SoundTouch library version string
 const char *SoundTouch::getVersionString()
 {
@@ -146,18 +137,14 @@ uint SoundTouch::getVersionId()
 // Sets the number of channels, 1 = mono, 2 = stereo
 void SoundTouch::setChannels(uint numChannels)
 {
-    /*if (numChannels != 1 && numChannels != 2) 
+    if (!verifyNumberOfChannels(numChannels)) return;
-    {
+
-        //ST_THROW_RT_ERROR("Illegal number of channels");
-		return;
-    }*/
     channels = numChannels;
     pRateTransposer->setChannels((int)numChannels);
     pTDStretch->setChannels((int)numChannels);
 }
 
 
-
 // Sets new rate control value. Normal rate = 1.0, smaller values
 // represent slower rate, larger faster rates.
 void SoundTouch::setRate(double newRate)
@@ -167,7 +154,6 @@ void SoundTouch::setRate(double newRate)
 }
 
 
-
 // Sets new rate control value as a difference in percents compared
 // to the original rate (-50 .. +100 %)
 void SoundTouch::setRateChange(double newRate)
@@ -177,7 +163,6 @@ void SoundTouch::setRateChange(double newRate)
 }
 
 
-
 // Sets new tempo control value. Normal tempo = 1.0, smaller values
 // represent slower tempo, larger faster tempo.
 void SoundTouch::setTempo(double newTempo)
@@ -187,7 +172,6 @@ void SoundTouch::setTempo(double newTempo)
 }
 
 
-
 // Sets new tempo control value as a difference in percents compared
 // to the original tempo (-50 .. +100 %)
 void SoundTouch::setTempoChange(double newTempo)
@@ -197,7 +181,6 @@ void SoundTouch::setTempoChange(double newTempo)
 }
 
 
-
 // Sets new pitch control value. Original pitch = 1.0, smaller values
 // represent lower pitches, larger values higher pitch.
 void SoundTouch::setPitch(double newPitch)
@@ -207,7 +190,6 @@ void SoundTouch::setPitch(double newPitch)
 }
 
 
-
 // Sets pitch change in octaves compared to the original pitch
 // (-1.00 .. +1.00)
 void SoundTouch::setPitchOctaves(double newPitch)
@@ -217,7 +199,6 @@ void SoundTouch::setPitchOctaves(double newPitch)
 }
 
 
-
 // Sets pitch change in semi-tones compared to the original pitch
 // (-12 .. +12)
 void SoundTouch::setPitchSemiTones(int newPitch)
@@ -226,7 +207,6 @@ void SoundTouch::setPitchSemiTones(int newPitch)
 }
 
 
-
 void SoundTouch::setPitchSemiTones(double newPitch)
 {
     setPitchOctaves(newPitch / 12.0);
@@ -286,9 +266,9 @@ void SoundTouch::calcEffectiveRateAndTempo()
 // Sets sample rate.
 void SoundTouch::setSampleRate(uint srate)
 {
-    bSrateSet = true;
     // set sample rate, leave other tempo changer parameters as they are.
     pTDStretch->setParameters((int)srate);
+    bSrateSet = true;
 }
 
 
@@ -305,22 +285,6 @@ void SoundTouch::putSamples(const SAMPLETYPE *samples, uint nSamples)
         ST_THROW_RT_ERROR("SoundTouch : Number of channels not defined");
     }
 
-    // Transpose the rate of the new samples if necessary
-    /* Bypass the nominal setting - can introduce a click in sound when tempo/pitch control crosses the nominal value...
-    if (rate == 1.0f) 
-    {
-        // The rate value is same as the original, simply evaluate the tempo changer. 
-        assert(output == pTDStretch);
-        if (pRateTransposer->isEmpty() == 0) 
-        {
-            // yet flush the last samples in the pitch transposer buffer
-            // (may happen if 'rate' changes from a non-zero value to zero)
-            pTDStretch->moveSamples(*pRateTransposer);
-        }
-        pTDStretch->putSamples(samples, nSamples);
-    } 
-    */
-
     // accumulate how many samples are expected out from processing, given the current 
     // processing setting
     samplesExpectedOut += (double)nSamples / ((double)rate * (double)tempo);
@@ -359,6 +323,7 @@ void SoundTouch::flush()
 
     // how many samples are still expected to output
     numStillExpected = (int)((long)(samplesExpectedOut + 0.5) - samplesOutput);
+    if (numStillExpected < 0) numStillExpected = 0;
 
     memset(buff, 0, 128 * channels * sizeof(SAMPLETYPE));
     // "Push" the last active samples out from the processing pipeline by
@@ -375,7 +340,6 @@ void SoundTouch::flush()
     delete[] buff;
 
     // Clear input buffers
- //   pRateTransposer->clearInput();
     pTDStretch->clearInput();
     // yet leave the output intouched as that's where the
     // flushed samples are!
@@ -446,25 +410,65 @@ int SoundTouch::getSetting(int settingId) const
             return pRateTransposer->getAAFilter()->getLength();
 
         case SETTING_USE_QUICKSEEK :
-            return (uint)   pTDStretch->isQuickSeekEnabled();
+            return (uint)pTDStretch->isQuickSeekEnabled();
 
         case SETTING_SEQUENCE_MS:
-            pTDStretch->getParameters(NULL, &temp, NULL, NULL);
+            pTDStretch->getParameters(nullptr, &temp, nullptr, nullptr);
             return temp;
 
         case SETTING_SEEKWINDOW_MS:
-            pTDStretch->getParameters(NULL, NULL, &temp, NULL);
+            pTDStretch->getParameters(nullptr, nullptr, &temp, nullptr);
             return temp;
 
         case SETTING_OVERLAP_MS:
-            pTDStretch->getParameters(NULL, NULL, NULL, &temp);
+            pTDStretch->getParameters(nullptr, nullptr, nullptr, &temp);
             return temp;
 
         case SETTING_NOMINAL_INPUT_SEQUENCE :
-			return pTDStretch->getInputSampleReq();
+        {
+            int size = pTDStretch->getInputSampleReq();
+
+#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
+            if (rate <= 1.0)
+            {
+                // transposing done before timestretch, which impacts latency
+                return (int)(size * rate + 0.5);
+            }
+#endif
+            return size;
+        }
 
         case SETTING_NOMINAL_OUTPUT_SEQUENCE :
-			return pTDStretch->getOutputBatchSize();
+        {
+            int size = pTDStretch->getOutputBatchSize();
+
+            if (rate > 1.0)
+            {
+                // transposing done after timestretch, which impacts latency
+                return (int)(size / rate + 0.5);
+            }
+            return size;
+        }
+
+        case SETTING_INITIAL_LATENCY:
+        {
+            double latency = pTDStretch->getLatency();
+            int latency_tr = pRateTransposer->getLatency();
+
+#ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
+            if (rate <= 1.0)
+            {
+                // transposing done before timestretch, which impacts latency
+                latency = (latency + latency_tr) * rate;
+            }
+            else
+#endif
+            {
+                latency += (double)latency_tr / rate;
+            }
+
+            return (int)(latency + 0.5);
+        }
 
         default :
             return 0;
@@ -477,12 +481,12 @@ int SoundTouch::getSetting(int settingId) const
 void SoundTouch::clear()
 {
     samplesExpectedOut = 0;
+    samplesOutput = 0;
     pRateTransposer->clear();
     pTDStretch->clear();
 }
 
 
-
 /// Returns number of samples currently unprocessed.
 uint SoundTouch::numUnprocessedSamples() const
 {
@@ -499,7 +503,6 @@ uint SoundTouch::numUnprocessedSamples() const
 }
 
 
-
 /// Output samples from beginning of the sample buffer. Copies requested samples to 
 /// output buffer and removes them from the sample buffer. If there are less than 
 /// 'numsample' samples in the buffer, returns all that available.
@@ -524,3 +527,12 @@ uint SoundTouch::receiveSamples(uint maxSamples)
     samplesOutput += (long)ret;
     return ret;
 }
+
+
+/// Get ratio between input and output audio durations, useful for calculating
+/// processed output duration: if you'll process a stream of N samples, then 
+/// you can expect to get out N * getInputOutputSampleRatio() samples.
+double SoundTouch::getInputOutputSampleRatio()
+{
+    return 1.0 / (tempo * rate);
+}

Externals/soundtouch/SoundTouch.h

@@ -41,13 +41,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-09-20 10:38:32 +0300 (Sun, 20 Sep 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: SoundTouch.h 230 2015-09-20 07:38:32Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -79,10 +72,10 @@ namespace soundtouch
 {
 
 /// Soundtouch library version string
-#define SOUNDTOUCH_VERSION          "1.9.2"
+#define SOUNDTOUCH_VERSION          "2.3.2"
 
 /// SoundTouch library version id
-#define SOUNDTOUCH_VERSION_ID       (10902)
+#define SOUNDTOUCH_VERSION_ID       (20302)
 
 //
 // Available setting IDs for the 'setSetting' & 'get_setting' functions:
@@ -116,15 +109,17 @@ namespace soundtouch
 #define SETTING_OVERLAP_MS          5
 
 
-/// Call "getSetting" with this ID to query nominal average processing sequence
+/// Call "getSetting" with this ID to query processing sequence size in samples.
-/// size in samples. This value tells approcimate value how many input samples 
+/// This value gives approximate value of how many input samples you'll need to
-/// SoundTouch needs to gather before it does DSP processing run for the sample batch.
+/// feed into SoundTouch after initial buffering to get out a new batch of
+/// output samples.
+///
+/// This value does not include initial buffering at beginning of a new processing
+/// stream, use SETTING_INITIAL_LATENCY to get the initial buffering size.
 ///
 /// Notices:
 /// - This is read-only parameter, i.e. setSetting ignores this parameter
-/// - Returned value is approximate average value, exact processing batch
+/// - This parameter value is not constant but change depending on
-///   size may wary from time to time
-/// - This parameter value is not constant but may change depending on 
 ///   tempo/pitch/rate/samplerate settings.
 #define SETTING_NOMINAL_INPUT_SEQUENCE      6
 
@@ -135,12 +130,41 @@ namespace soundtouch
 ///
 /// Notices:
 /// - This is read-only parameter, i.e. setSetting ignores this parameter
-/// - Returned value is approximate average value, exact processing batch
+/// - This parameter value is not constant but change depending on
-///   size may wary from time to time
-/// - This parameter value is not constant but may change depending on 
 ///   tempo/pitch/rate/samplerate settings.
 #define SETTING_NOMINAL_OUTPUT_SEQUENCE     7
 
+
+/// Call "getSetting" with this ID to query initial processing latency, i.e.
+/// approx. how many samples you'll need to enter to SoundTouch pipeline before
+/// you can expect to get first batch of ready output samples out.
+///
+/// After the first output batch, you can then expect to get approx.
+/// SETTING_NOMINAL_OUTPUT_SEQUENCE ready samples out for every
+/// SETTING_NOMINAL_INPUT_SEQUENCE samples that you enter into SoundTouch.
+///
+/// Example:
+///     processing with parameter -tempo=5
+///     => initial latency = 5509 samples
+///        input sequence  = 4167 samples
+///        output sequence = 3969 samples
+///
+/// Accordingly, you can expect to feed in approx. 5509 samples at beginning of
+/// the stream, and then you'll get out the first 3969 samples. After that, for
+/// every approx. 4167 samples that you'll put in, you'll receive again approx.
+/// 3969 samples out.
+///
+/// This also means that average latency during stream processing is
+/// INITIAL_LATENCY-OUTPUT_SEQUENCE/2, in the above example case 5509-3969/2
+/// = 3524 samples
+///
+/// Notices:
+/// - This is read-only parameter, i.e. setSetting ignores this parameter
+/// - This parameter value is not constant but change depending on
+///   tempo/pitch/rate/samplerate settings.
+#define SETTING_INITIAL_LATENCY             8
+
+
 class SoundTouch : public FIFOProcessor
 {
 private:
@@ -185,7 +209,7 @@ protected :
 
 public:
     SoundTouch();
-    virtual ~SoundTouch();
+    virtual ~SoundTouch() override;
 
     /// Get SoundTouch library version string
     static const char *getVersionString();
@@ -228,6 +252,24 @@ public:
     /// Sets sample rate.
     void setSampleRate(uint srate);
 
+    /// Get ratio between input and output audio durations, useful for calculating
+    /// processed output duration: if you'll process a stream of N samples, then
+    /// you can expect to get out N * getInputOutputSampleRatio() samples.
+    ///
+    /// This ratio will give accurate target duration ratio for a full audio track,
+    /// given that the the whole track is processed with same processing parameters.
+    ///
+    /// If this ratio is applied to calculate intermediate offsets inside a processing
+    /// stream, then this ratio is approximate and can deviate +- some tens of milliseconds
+    /// from ideal offset, yet by end of the audio stream the duration ratio will become
+    /// exact.
+    ///
+    /// Example: if processing with parameters "-tempo=15 -pitch=-3", the function
+    /// will return value 0.8695652... Now, if processing an audio stream whose duration
+    /// is exactly one million audio samples, then you can expect the processed
+    /// output duration  be 0.869565 * 1000000 = 869565 samples.
+    double getInputOutputSampleRatio();
+
     /// Flushes the last samples from the processing pipeline to the output.
     /// Clears also the internal processing buffers.
     //
@@ -245,7 +287,7 @@ public:
             uint numSamples                         ///< Number of samples in buffer. Notice
                                                     ///< that in case of stereo-sound a single sample
                                                     ///< contains data for both channels.
-            );
+            ) override;
 
     /// Output samples from beginning of the sample buffer. Copies requested samples to
     /// output buffer and removes them from the sample buffer. If there are less than
@@ -254,7 +296,7 @@ public:
     /// \return Number of samples returned.
     virtual uint receiveSamples(SAMPLETYPE *output, ///< Buffer where to copy output samples.
         uint maxSamples                 ///< How many samples to receive at max.
-        );
+        ) override;
 
     /// Adjusts book-keeping so that given number of samples are removed from beginning of the
     /// sample buffer without copying them anywhere.
@@ -262,16 +304,16 @@ public:
     /// Used to reduce the number of samples in the buffer when accessing the sample buffer directly
     /// with 'ptrBegin' function.
     virtual uint receiveSamples(uint maxSamples   ///< Remove this many samples from the beginning of pipe.
-        );
+        ) override;
 
     /// Clears all the samples in the object's output and internal processing
     /// buffers.
-    virtual void clear();
+    virtual void clear() override;
 
     /// Changes a setting controlling the processing system behaviour. See the
     /// 'SETTING_...' defines for available setting ID's.
     ///
-    /// \return 'true' if the setting was succesfully changed
+    /// \return 'true' if the setting was successfully changed
     bool setSetting(int settingId,   ///< Setting ID number. see SETTING_... defines.
                     int value        ///< New setting value.
                     );
@@ -286,6 +328,11 @@ public:
     /// Returns number of samples currently unprocessed.
     virtual uint numUnprocessedSamples() const;
 
+    /// Return number of channels
+    uint numChannels() const
+    {
+        return channels;
+    }
 
     /// Other handy functions that are implemented in the ancestor classes (see
     /// classes 'FIFOProcessor' and 'FIFOSamplePipe')

Externals/soundtouch/TDStretch.cpp

@@ -1,11 +1,17 @@
-////////////////////////////////////////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////////
 /// 
 /// Sampled sound tempo changer/time stretch algorithm. Changes the sound tempo 
 /// while maintaining the original pitch by using a time domain WSOLA-like 
 /// method with several performance-increasing tweaks.
 ///
-/// Note : MMX optimized functions reside in a separate, platform-specific 
+/// Notes : MMX optimized functions reside in a separate, platform-specific 
-/// file, e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp'
+/// file, e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp'.
+///
+/// This source file contains OpenMP optimizations that allow speeding up the
+/// corss-correlation algorithm by executing it in several threads / CPU cores 
+/// in parallel. See the following article link for more detailed discussion 
+/// about SoundTouch OpenMP optimizations:
+/// http://www.softwarecoven.com/parallel-computing-in-embedded-mobile-devices
 ///
 /// Author        : Copyright (c) Olli Parviainen
 /// Author e-mail : oparviai 'at' iki.fi
@@ -13,13 +19,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
-// File revision : $Revision: 1.12 $
-//
-// $Id: TDStretch.cpp 226 2015-08-08 21:00:15Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -55,26 +54,6 @@ using namespace soundtouch;
 
 #define max(x, y) (((x) > (y)) ? (x) : (y))
 
-
-/*****************************************************************************
- *
- * Constant definitions
- *
- *****************************************************************************/
-
-// Table for the hierarchical mixing position seeking algorithm
-const short _scanOffsets[5][24]={
-    { 124,  186,  248,  310,  372,  434,  496,  558,  620,  682,  744, 806,
-      868,  930,  992, 1054, 1116, 1178, 1240, 1302, 1364, 1426, 1488,   0},
-    {-100,  -75,  -50,  -25,   25,   50,   75,  100,    0,    0,    0,   0,
-        0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,   0},
-    { -20,  -15,  -10,   -5,    5,   10,   15,   20,    0,    0,    0,   0,
-        0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,   0},
-    {  -4,   -3,   -2,   -1,    1,    2,    3,    4,    0,    0,    0,   0,
-        0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,   0},
-    { 121,  114,   97,  114,   98,  105,  108,   32,  104,   99,  117,  111,
-      116,  100,  110,  117,  111,  115,    0,    0,    0,    0,    0,   0}};
-
 /*****************************************************************************
  *
  * Implementation of the class 'TDStretch'
@@ -87,18 +66,13 @@ TDStretch::TDStretch() : FIFOProcessor(&outputBuffer)
     bQuickSeek = false;
     channels = 2;
 
-    pMidBuffer = NULL;
+    pMidBuffer = nullptr;
-    pMidBufferUnaligned = NULL;
+    pMidBufferUnaligned = nullptr;
     overlapLength = 0;
 
     bAutoSeqSetting = true;
     bAutoSeekSetting = true;
 
-    maxnorm = 0;
-    maxnormf = 1e8;
-
-    skipFract = 0;
-
     tempo = 1.0f;
     setParameters(44100, DEFAULT_SEQUENCE_MS, DEFAULT_SEEKWINDOW_MS, DEFAULT_OVERLAP_MS);
     setTempo(1.0f);
@@ -128,7 +102,12 @@ void TDStretch::setParameters(int aSampleRate, int aSequenceMS,
                               int aSeekWindowMS, int aOverlapMS)
 {
     // accept only positive parameter values - if zero or negative, use old values instead
-    if (aSampleRate > 0)   this->sampleRate = aSampleRate;
+    if (aSampleRate > 0)
+    {
+        if (aSampleRate > 192000) ST_THROW_RT_ERROR("Error: Excessive samplerate");
+        this->sampleRate = aSampleRate;
+    }
+
     if (aOverlapMS > 0) this->overlapMs = aOverlapMS;
 
     if (aSequenceMS > 0)
@@ -164,7 +143,7 @@ void TDStretch::setParameters(int aSampleRate, int aSequenceMS,
 
 
 /// Get routine control parameters, see setParameters() function.
-/// Any of the parameters to this function can be NULL, in such case corresponding parameter
+/// Any of the parameters to this function can be nullptr, in such case corresponding parameter
 /// value isn't returned.
 void TDStretch::getParameters(int *pSampleRate, int *pSequenceMs, int *pSeekWindowMs, int *pOverlapMs) const
 {
@@ -219,6 +198,10 @@ void TDStretch::clearInput()
 {
     inputBuffer.clear();
     clearMidBuffer();
+    isBeginning = true;
+    maxnorm = 0;
+    maxnormf = 1e8;
+    skipFract = 0;
 }
 
 
@@ -297,21 +280,23 @@ int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos)
     int i;
     double norm;
 
-    bestCorr = FLT_MIN;
+    bestCorr = -FLT_MAX;
     bestOffs = 0;
 
     // Scans for the best correlation value by testing each possible position
     // over the permitted range.
     bestCorr = calcCrossCorr(refPos, pMidBuffer, norm);
+    bestCorr = (bestCorr + 0.1) * 0.75;
 
     #pragma omp parallel for
     for (i = 1; i < seekLength; i ++)
     {
         double corr;
         // Calculates correlation value for the mixing position corresponding to 'i'
-#ifdef _OPENMP
+#if defined(_OPENMP) || defined(ST_SIMD_AVOID_UNALIGNED)
         // in parallel OpenMP mode, can't use norm accumulator version as parallel executor won't
         // iterate the loop in sequential order
+        // in SIMD mode, avoid accumulator version to allow avoiding unaligned positions
         corr = calcCrossCorr(refPos + channels * i, pMidBuffer, norm);
 #else
         // In non-parallel version call "calcCrossCorrAccumulate" that is otherwise same
@@ -354,7 +339,7 @@ int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos)
 // with improved precision
 //
 // Based on testing:
-// - This algorithm gives on average 99% as good match as the full algorith
+// - This algorithm gives on average 99% as good match as the full algorithm
 // - this quick seek algorithm finds the best match on ~90% of cases
 // - on those 10% of cases when this algorithm doesn't find best match, 
 //   it still finds on average ~90% match vs. the best possible match
@@ -373,12 +358,10 @@ int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos)
 
     // note: 'float' types used in this function in case that the platform would need to use software-fp
 
-    bestCorr = FLT_MIN;
+    bestCorr =
-    bestOffs = SCANWIND;
+    bestCorr2 = -FLT_MAX;
-    bestCorr2 = FLT_MIN;
+    bestOffs = 
-    bestOffs2 = 0;
+    bestOffs2 = SCANWIND;
-
-    int best = 0;
 
     // Scans for the best correlation value by testing each possible position
     // over the permitted range. Look for two best matches on the first pass to
@@ -436,7 +419,6 @@ int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos)
         {
             bestCorr = corr;
             bestOffs = i;
-            best = 1;
         }
     }
 
@@ -458,7 +440,6 @@ int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos)
         {
             bestCorr = corr;
             bestOffs = i;
-            best = 2;
         }
     }
 
@@ -515,18 +496,18 @@ void TDStretch::clearCrossCorrState()
 void TDStretch::calcSeqParameters()
 {
     // Adjust tempo param according to tempo, so that variating processing sequence length is used
-    // at varius tempo settings, between the given low...top limits
+    // at various tempo settings, between the given low...top limits
     #define AUTOSEQ_TEMPO_LOW   0.5     // auto setting low tempo range (-50%)
     #define AUTOSEQ_TEMPO_TOP   2.0     // auto setting top tempo range (+100%)
 
     // sequence-ms setting values at above low & top tempo
-    #define AUTOSEQ_AT_MIN      125.0
+    #define AUTOSEQ_AT_MIN      90.0
-    #define AUTOSEQ_AT_MAX      50.0
+    #define AUTOSEQ_AT_MAX      40.0
     #define AUTOSEQ_K           ((AUTOSEQ_AT_MAX - AUTOSEQ_AT_MIN) / (AUTOSEQ_TEMPO_TOP - AUTOSEQ_TEMPO_LOW))
     #define AUTOSEQ_C           (AUTOSEQ_AT_MIN - (AUTOSEQ_K) * (AUTOSEQ_TEMPO_LOW))
 
     // seek-window-ms setting values at above low & top tempoq
-    #define AUTOSEEK_AT_MIN     25.0
+    #define AUTOSEEK_AT_MIN     20.0
     #define AUTOSEEK_AT_MAX     15.0
     #define AUTOSEEK_K          ((AUTOSEEK_AT_MAX - AUTOSEEK_AT_MIN) / (AUTOSEQ_TEMPO_TOP - AUTOSEQ_TEMPO_LOW))
     #define AUTOSEEK_C          (AUTOSEEK_AT_MIN - (AUTOSEEK_K) * (AUTOSEQ_TEMPO_LOW))
@@ -586,9 +567,8 @@ void TDStretch::setTempo(double newTempo)
 // Sets the number of channels, 1 = mono, 2 = stereo
 void TDStretch::setChannels(int numChannels)
 {
-    assert(numChannels > 0);
+    if (!verifyNumberOfChannels(numChannels) ||
-    if (channels == numChannels) return;
+        (channels == numChannels)) return;
-//    assert(numChannels == 1 || numChannels == 2);
 
     channels = numChannels;
     inputBuffer.setChannels(channels);
@@ -637,7 +617,8 @@ void TDStretch::processNominalTempo()
 // the result into 'outputBuffer'
 void TDStretch::processSamples()
 {
-    int ovlSkip, offset;
+    int ovlSkip;
+    int offset = 0;
     int temp;
 
     /* Removed this small optimization - can introduce a click to sound when tempo setting
@@ -654,8 +635,10 @@ void TDStretch::processSamples()
     // to form a processing frame.
     while ((int)inputBuffer.numSamples() >= sampleReq) 
     {
-        // If tempo differs from the normal ('SCALE'), scan for the best overlapping
+        if (isBeginning == false)
-        // position
+        {
+            // apart from the very beginning of the track, 
+            // scan for the best overlapping position & do overlap-add
             offset = seekBestOverlapPosition(inputBuffer.ptrBegin());
 
             // Mix the samples in the 'inputBuffer' at position of 'offset' with the 
@@ -664,25 +647,50 @@ void TDStretch::processSamples()
             // (that's in 'midBuffer')
             overlap(outputBuffer.ptrEnd((uint)overlapLength), inputBuffer.ptrBegin(), (uint)offset);
             outputBuffer.putSamples((uint)overlapLength);
+            offset += overlapLength;
+        }
+        else
+        {
+            // Adjust processing offset at beginning of track by not perform initial overlapping
+            // and compensating that in the 'input buffer skip' calculation
+            isBeginning = false;
+            int skip = (int)(tempo * overlapLength + 0.5 * seekLength + 0.5);
+
+            #ifdef ST_SIMD_AVOID_UNALIGNED
+            // in SIMD mode, round the skip amount to value corresponding to aligned memory address
+            if (channels == 1)
+            {
+                skip &= -4;
+            }
+            else if (channels == 2)
+            {
+                skip &= -2;
+            }
+            #endif
+            skipFract -= skip;
+            if (skipFract <= -nominalSkip)
+            {
+                skipFract = -nominalSkip;
+            }
+        }
 
         // ... then copy sequence samples from 'inputBuffer' to output:
 
-        // length of sequence
-        temp = (seekWindowLength - 2 * overlapLength);
-
         // crosscheck that we don't have buffer overflow...
-        if ((int)inputBuffer.numSamples() < (offset + temp + overlapLength * 2))
+        if ((int)inputBuffer.numSamples() < (offset + seekWindowLength - overlapLength))
         {
             continue;    // just in case, shouldn't really happen
         }
 
-        outputBuffer.putSamples(inputBuffer.ptrBegin() + channels * (offset + overlapLength), (uint)temp);
+        // length of sequence
+        temp = (seekWindowLength - 2 * overlapLength);
+        outputBuffer.putSamples(inputBuffer.ptrBegin() + channels * offset, (uint)temp);
 
         // Copies the end of the current sequence from 'inputBuffer' to 
         // 'midBuffer' for being mixed with the beginning of the next 
         // processing sequence and so on
-        assert((offset + temp + overlapLength * 2) <= (int)inputBuffer.numSamples());
+        assert((offset + temp + overlapLength) <= (int)inputBuffer.numSamples());
-        memcpy(pMidBuffer, inputBuffer.ptrBegin() + channels * (offset + temp + overlapLength), 
+        memcpy(pMidBuffer, inputBuffer.ptrBegin() + channels * (offset + temp), 
             channels * sizeof(SAMPLETYPE) * overlapLength);
 
         // Remove the processed samples from the input buffer. Update
@@ -732,7 +740,7 @@ void TDStretch::acceptNewOverlapLength(int newOverlapLength)
 
 // Operator 'new' is overloaded so that it automatically creates a suitable instance 
 // depending on if we've a MMX/SSE/etc-capable CPU available or not.
-void * TDStretch::operator new(size_t s)
+void * TDStretch::operator new(size_t)
 {
     // Notice! don't use "new TDStretch" directly, use "newInstance" to create a new instance instead!
     ST_THROW_RT_ERROR("Error in TDStretch::new: Don't use 'new TDStretch' directly, use 'newInstance' member instead!");
@@ -776,7 +784,7 @@ TDStretch * TDStretch::newInstance()
 
 //////////////////////////////////////////////////////////////////////////////
 //
-// Integer arithmetics specific algorithm implementations.
+// Integer arithmetic specific algorithm implementations.
 //
 //////////////////////////////////////////////////////////////////////////////
 
@@ -802,21 +810,19 @@ void TDStretch::overlapStereo(short *poutput, const short *input) const
 
 // Overlaps samples in 'midBuffer' with the samples in 'input'. The 'Multi'
 // version of the routine.
-void TDStretch::overlapMulti(SAMPLETYPE *poutput, const SAMPLETYPE *input) const
+void TDStretch::overlapMulti(short *poutput, const short *input) const
 {
-    SAMPLETYPE m1=(SAMPLETYPE)0;
+    short m1;
-    SAMPLETYPE m2;
+    int i = 0;
-    int i=0;
 
-    for (m2 = (SAMPLETYPE)overlapLength; m2; m2 --)
+    for (m1 = 0; m1 < overlapLength; m1 ++)
     {
+        short m2 = (short)(overlapLength - m1);
         for (int c = 0; c < channels; c ++)
         {
             poutput[i] = (input[i] * m1 + pMidBuffer[i] * m2)  / overlapLength;
             i++;
         }
-
-        m1++;
     }
 }
 
@@ -861,26 +867,34 @@ double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, do
     unsigned long lnorm;
     int i;
 
+    #ifdef ST_SIMD_AVOID_UNALIGNED
+        // in SIMD mode skip 'mixingPos' positions that aren't aligned to 16-byte boundary
+        if (((ulongptr)mixingPos) & 15) return -1e50;
+    #endif
+
+    // hint compiler autovectorization that loop length is divisible by 8
+    int ilength = (channels * overlapLength) & -8;
+
     corr = lnorm = 0;
-    // Same routine for stereo and mono. For stereo, unroll loop for better
+    // Same routine for stereo and mono
-    // efficiency and gives slightly better resolution against rounding. 
+    for (i = 0; i < ilength; i += 2)
-    // For mono it same routine, just  unrolls loop by factor of 4
-    for (i = 0; i < channels * overlapLength; i += 4) 
     {
         corr += (mixingPos[i] * compare[i] + 
-                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;  // notice: do intermediate division here to avoid integer overflow
+                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;
-        corr += (mixingPos[i + 2] * compare[i + 2] + 
-                mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm;
         lnorm += (mixingPos[i] * mixingPos[i] + 
-                mixingPos[i + 1] * mixingPos[i + 1]) >> overlapDividerBitsNorm; // notice: do intermediate division here to avoid integer overflow
+                  mixingPos[i + 1] * mixingPos[i + 1]) >> overlapDividerBitsNorm;
-        lnorm += (mixingPos[i + 2] * mixingPos[i + 2] + 
+        // do intermediate scalings to avoid integer overflow
-                mixingPos[i + 3] * mixingPos[i + 3]) >> overlapDividerBitsNorm;
     }
 
+    if (lnorm > maxnorm)
+    {
+        // modify 'maxnorm' inside critical section to avoid multi-access conflict if in OpenMP mode
+        #pragma omp critical
         if (lnorm > maxnorm)
         {
             maxnorm = lnorm;
         }
+    }
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
     norm = (double)lnorm;
@@ -892,9 +906,12 @@ double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, do
 double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm)
 {
     long corr;
-    unsigned long lnorm;
+    long lnorm;
     int i;
 
+    // hint compiler autovectorization that loop length is divisible by 8
+    int ilength = (channels * overlapLength) & -8;
+
     // cancel first normalizer tap from previous round
     lnorm = 0;
     for (i = 1; i <= channels; i ++)
@@ -903,15 +920,11 @@ double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *c
     }
 
     corr = 0;
-    // Same routine for stereo and mono. For stereo, unroll loop for better
+    // Same routine for stereo and mono.
-    // efficiency and gives slightly better resolution against rounding. 
+    for (i = 0; i < ilength; i += 2) 
-    // For mono it same routine, just  unrolls loop by factor of 4
-    for (i = 0; i < channels * overlapLength; i += 4) 
     {
         corr += (mixingPos[i] * compare[i] + 
-                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;  // notice: do intermediate division here to avoid integer overflow
+                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;
-        corr += (mixingPos[i + 2] * compare[i + 2] + 
-                 mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm;
     }
 
     // update normalizer with last samples of this round
@@ -936,7 +949,7 @@ double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *c
 
 //////////////////////////////////////////////////////////////////////////////
 //
-// Floating point arithmetics specific algorithm implementations.
+// Floating point arithmetic specific algorithm implementations.
 //
 
 #ifdef SOUNDTOUCH_FLOAT_SAMPLES
@@ -1012,27 +1025,24 @@ void TDStretch::calculateOverlapLength(int overlapInMsec)
 /// Calculate cross-correlation
 double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, double &anorm)
 {
-    double corr;
+    float corr;
-    double norm;
+    float norm;
     int i;
 
+    #ifdef ST_SIMD_AVOID_UNALIGNED
+        // in SIMD mode skip 'mixingPos' positions that aren't aligned to 16-byte boundary
+        if (((ulongptr)mixingPos) & 15) return -1e50;
+    #endif
+
+    // hint compiler autovectorization that loop length is divisible by 8
+    int ilength = (channels * overlapLength) & -8;
+
     corr = norm = 0;
-    // Same routine for stereo and mono. For Stereo, unroll by factor of 2.
+    // Same routine for stereo and mono
-    // For mono it's same routine yet unrollsd by factor of 4.
+    for (i = 0; i < ilength; i ++)
-    for (i = 0; i < channels * overlapLength; i += 4) 
     {
-        corr += mixingPos[i] * compare[i] +
+        corr += mixingPos[i] * compare[i];
-                mixingPos[i + 1] * compare[i + 1];
+        norm += mixingPos[i] * mixingPos[i];
-
-        norm += mixingPos[i] * mixingPos[i] + 
-                mixingPos[i + 1] * mixingPos[i + 1];
-
-        // unroll the loop for better CPU efficiency:
-        corr += mixingPos[i + 2] * compare[i + 2] +
-                mixingPos[i + 3] * compare[i + 3];
-
-        norm += mixingPos[i + 2] * mixingPos[i + 2] +
-                mixingPos[i + 3] * mixingPos[i + 3];
     }
 
     anorm = norm;
@@ -1043,7 +1053,7 @@ double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, do
 /// Update cross-correlation by accumulating "norm" coefficient by previously calculated value
 double TDStretch::calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm)
 {
-    double corr;
+    float corr;
     int i;
 
     corr = 0;
@@ -1054,14 +1064,13 @@ double TDStretch::calcCrossCorrAccumulate(const float *mixingPos, const float *c
         norm -= mixingPos[-i] * mixingPos[-i];
     }
 
-    // Same routine for stereo and mono. For Stereo, unroll by factor of 2.
+    // hint compiler autovectorization that loop length is divisible by 8
-    // For mono it's same routine yet unrollsd by factor of 4.
+    int ilength = (channels * overlapLength) & -8;
-    for (i = 0; i < channels * overlapLength; i += 4) 
+
+    // Same routine for stereo and mono
+    for (i = 0; i < ilength; i ++)
     {
-        corr += mixingPos[i] * compare[i] +
+        corr += mixingPos[i] * compare[i];
-                mixingPos[i + 1] * compare[i + 1] +
-                mixingPos[i + 2] * compare[i + 2] +
-                mixingPos[i + 3] * compare[i + 3];
     }
 
     // update normalizer with last samples of this round

Externals/soundtouch/TDStretch.h

@@ -13,13 +13,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: TDStretch.h 226 2015-08-08 21:00:15Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -134,6 +127,7 @@ protected:
     bool bQuickSeek;
     bool bAutoSeqSetting;
     bool bAutoSeekSetting;
+    bool isBeginning;
 
     SAMPLETYPE *pMidBuffer;
     SAMPLETYPE *pMidBufferUnaligned;
@@ -163,7 +157,6 @@ protected:
     void calcSeqParameters();
     void adaptNormalizer();
 
-
     /// Changes the tempo of the given sound samples.
     /// Returns amount of samples returned in the "output" buffer.
     /// The maximum amount of samples that can be returned at a time is set by
@@ -172,7 +165,7 @@ protected:
     
 public:
     TDStretch();
-    virtual ~TDStretch();
+    virtual ~TDStretch() override;
 
     /// Operator 'new' is overloaded so that it automatically creates a suitable instance 
     /// depending on if we've a MMX/SSE/etc-capable CPU available or not.
@@ -194,7 +187,7 @@ public:
     void setTempo(double newTempo);
 
     /// Returns nonzero if there aren't any samples available for outputting.
-    virtual void clear();
+    virtual void clear() override;
 
     /// Clears the input buffer
     void clearInput();
@@ -224,7 +217,7 @@ public:
                        );
 
     /// Get routine control parameters, see setParameters() function.
-    /// Any of the parameters to this function can be NULL, in such case corresponding parameter
+    /// Any of the parameters to this function can be nullptr, in such case corresponding parameter
     /// value isn't returned.
     void getParameters(int *pSampleRate, int *pSequenceMs, int *pSeekWindowMs, int *pOverlapMs) const;
 
@@ -234,7 +227,7 @@ public:
             const SAMPLETYPE *samples,  ///< Input sample data
             uint numSamples                         ///< Number of samples in 'samples' so that one sample
                                                     ///< contains both channels if stereo
-            );
+            ) override;
 
     /// return nominal input sample requirement for triggering a processing batch
     int getInputSampleReq() const
@@ -247,8 +240,13 @@ public:
     {
         return seekWindowLength - overlapLength;
     }
-};
 
+	/// return approximate initial input-output latency
+	int getLatency() const
+	{
+		return sampleReq;
+	}
+};
 
 
 // Implementation-specific class declarations:
@@ -258,10 +256,10 @@ public:
     class TDStretchMMX : public TDStretch
     {
     protected:
-        double calcCrossCorr(const short *mixingPos, const short *compare, double &norm);
+        double calcCrossCorr(const short *mixingPos, const short *compare, double &norm) override;
-        double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm);
+        double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm) override;
-        virtual void overlapStereo(short *output, const short *input) const;
+        virtual void overlapStereo(short *output, const short *input) const override;
-        virtual void clearCrossCorrState();
+        virtual void clearCrossCorrState() override;
     };
 #endif /// SOUNDTOUCH_ALLOW_MMX
 
@@ -271,8 +269,8 @@ public:
     class TDStretchSSE : public TDStretch
     {
     protected:
-        double calcCrossCorr(const float *mixingPos, const float *compare, double &norm);
+        double calcCrossCorr(const float *mixingPos, const float *compare, double &norm) override;
-        double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm);
+        double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm) override;
     };
 
 #endif /// SOUNDTOUCH_ALLOW_SSE

Externals/soundtouch/cpu_detect.h

@@ -12,13 +12,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2008-02-10 18:26:55 +0200 (Sun, 10 Feb 2008) $
-// File revision : $Revision: 4 $
-//
-// $Id: cpu_detect.h 11 2008-02-10 16:26:55Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -51,8 +44,6 @@
 #define SUPPORT_SSE         0x0008
 #define SUPPORT_SSE2        0x0010
 
-using namespace soundtouch;
-
 /// Checks which instruction set extensions are supported by the CPU.
 ///
 /// \return A bitmask of supported extensions, see SUPPORT_... defines.

Externals/soundtouch/cpu_detect_x86.cpp

@@ -11,13 +11,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-07 20:24:28 +0200 (Tue, 07 Jan 2014) $
-// File revision : $Revision: 4 $
-//
-// $Id: cpu_detect_x86.cpp 183 2014-01-07 18:24:28Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -75,7 +68,6 @@ void disableExtensions(uint dwDisableMask)
 }
 
 
-
 /// Checks which instruction set extensions are supported by the CPU.
 uint detectCPUextensions(void)
 {

Externals/soundtouch/mmx_optimized.cpp

@@ -20,13 +20,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: mmx_optimized.cpp 226 2015-08-08 21:00:15Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -123,10 +116,15 @@ double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &d
     // Clear MMS state
     _m_empty();
 
+    if (norm > (long)maxnorm)
+    {
+        // modify 'maxnorm' inside critical section to avoid multi-access conflict if in OpenMP mode
+        #pragma omp critical
         if (norm > (long)maxnorm)
         {
             maxnorm = norm;
         }
+    }
 
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
@@ -219,7 +217,6 @@ void TDStretchMMX::clearCrossCorrState()
 }
 
 
-
 // MMX-optimized version of the function overlapStereo
 void TDStretchMMX::overlapStereo(short *output, const short *input) const
 {
@@ -297,8 +294,8 @@ void TDStretchMMX::overlapStereo(short *output, const short *input) const
 
 FIRFilterMMX::FIRFilterMMX() : FIRFilter()
 {
-    filterCoeffsAlign = NULL;
+    filterCoeffsAlign = nullptr;
-    filterCoeffsUnalign = NULL;
+    filterCoeffsUnalign = nullptr;
 }
 
 
@@ -335,7 +332,6 @@ void FIRFilterMMX::setCoefficients(const short *coeffs, uint newLength, uint uRe
 }
 
 
-
 // mmx-optimized version of the filter routine for stereo sound
 uint FIRFilterMMX::evaluateFilterStereo(short *dest, const short *src, uint numSamples) const
 {
@@ -392,4 +388,9 @@ uint FIRFilterMMX::evaluateFilterStereo(short *dest, const short *src, uint numS
     return (numSamples & 0xfffffffe) - length;
 }
 
+#else
+
+// workaround to not complain about empty module
+bool _dontcomplain_mmx_empty;
+
 #endif  // SOUNDTOUCH_ALLOW_MMX

Externals/soundtouch/sse_optimized.cpp

@@ -23,13 +23,6 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
-// File revision : $Revision: 4 $
-//
-// $Id: sse_optimized.cpp 226 2015-08-08 21:00:15Z oparviai $
-//
-////////////////////////////////////////////////////////////////////////////////
-//
 // License :
 //
 //  SoundTouch audio processing library
@@ -87,7 +80,7 @@ double TDStretchSSE::calcCrossCorr(const float *pV1, const float *pV2, double &a
     // Compile-time define SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION is provided
     // for choosing if this little cheating is allowed.
 
-#ifdef SOUNDTOUCH_ALLOW_NONEXACT_SIMD_OPTIMIZATION
+#ifdef ST_SIMD_AVOID_UNALIGNED
     // Little cheating allowed, return valid correlation only for 
     // aligned locations, meaning every second round for stereo sound.
 
@@ -202,16 +195,16 @@ double TDStretchSSE::calcCrossCorrAccumulate(const float *pV1, const float *pV2,
 
 FIRFilterSSE::FIRFilterSSE() : FIRFilter()
 {
-    filterCoeffsAlign = NULL;
+    filterCoeffsAlign = nullptr;
-    filterCoeffsUnalign = NULL;
+    filterCoeffsUnalign = nullptr;
 }
 
 
 FIRFilterSSE::~FIRFilterSSE()
 {
     delete[] filterCoeffsUnalign;
-    filterCoeffsAlign = NULL;
+    filterCoeffsAlign = nullptr;
-    filterCoeffsUnalign = NULL;
+    filterCoeffsUnalign = nullptr;
 }
 
 
@@ -252,10 +245,10 @@ uint FIRFilterSSE::evaluateFilterStereo(float *dest, const float *source, uint n
 
     if (count < 2) return 0;
 
-    assert(source != NULL);
+    assert(source != nullptr);
-    assert(dest != NULL);
+    assert(dest != nullptr);
     assert((length % 8) == 0);
-    assert(filterCoeffsAlign != NULL);
+    assert(filterCoeffsAlign != nullptr);
     assert(((ulongptr)filterCoeffsAlign) % 16 == 0);
 
     // filter is evaluated for two stereo samples with each iteration, thus use of 'j += 2'