Merge pull request #3406 from skidau/soundtouch-192

Updated SoundTouch library to 1.9.2

Externals/soundtouch/AAFilter.cpp

@@ -12,7 +12,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-06 08:40:22 +1100 (Mon, 06 Jan 2014) $
+// 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 $

Externals/soundtouch/AAFilter.h

@@ -13,7 +13,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-08 06:41:23 +1100 (Wed, 08 Jan 2014) $
+// 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 $

Externals/soundtouch/BPMDetect.cpp

@@ -26,10 +26,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-08-31 05:45:25 +1000 (Fri, 31 Aug 2012) $
+// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: BPMDetect.cpp 149 2012-08-30 19:45:25Z oparviai $
+// $Id: BPMDetect.cpp 202 2015-02-21 21:24:29Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -226,6 +226,7 @@ void BPMDetect::updateXCorr(int process_samples)
     assert(buffer->numSamples() >= (uint)(process_samples + windowLen));
 
     pBuffer = buffer->ptrBegin();
+    #pragma omp parallel for
     for (offs = windowStart; offs < windowLen; offs ++) 
     {
         LONG_SAMPLETYPE sum;

Externals/soundtouch/BPMDetect.h

@@ -26,7 +26,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-08-30 19:53:44 +0000 (Thu, 30 Aug 2012) $
+// 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 $

Externals/soundtouch/FIFOSampleBuffer.cpp

@@ -15,7 +15,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-11-09 05:53:01 +1100 (Fri, 09 Nov 2012) $
+// 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 $
@@ -86,10 +86,6 @@ void FIFOSampleBuffer::setChannels(int numChannels)
     samplesInBuffer = usedBytes / channels;
 }
 
-int FIFOSampleBuffer::getChannels()
-{
-	return channels;
-}
 
 // if output location pointer 'bufferPos' isn't zero, 'rewinds' the buffer and
 // zeroes this pointer by copying samples from the 'bufferPos' pointer 

Externals/soundtouch/FIFOSampleBuffer.h

@@ -15,10 +15,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-06-13 19:29:53 +0000 (Wed, 13 Jun 2012) $
+// Last changed  : $Date: 2014-01-05 23:40:22 +0200 (Sun, 05 Jan 2014) $
 // File revision : $Revision: 4 $
 //
-// $Id: FIFOSampleBuffer.h 143 2012-06-13 19:29:53Z oparviai $
+// $Id: FIFOSampleBuffer.h 177 2014-01-05 21:40:22Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -161,7 +161,12 @@ public:
 
     /// Sets number of channels, 1 = mono, 2 = stereo.
     void setChannels(int numChannels);
-	int getChannels();
+
+    /// Get number of channels
+    int getChannels() 
+    {
+        return channels;
+    }
 
     /// Returns nonzero if there aren't any samples available for outputting.
     virtual int isEmpty() const;

Externals/soundtouch/FIFOSamplePipe.h

@@ -17,7 +17,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-06-13 19:29:53 +0000 (Wed, 13 Jun 2012) $
+// 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 $

Externals/soundtouch/FIRFilter.cpp

@@ -11,10 +11,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2013-06-13 01:24:44 +1000 (Thu, 13 Jun 2013) $
+// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: FIRFilter.cpp 171 2013-06-12 15:24:44Z oparviai $
+// $Id: FIRFilter.cpp 202 2015-02-21 21:24:29Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -72,8 +72,7 @@ FIRFilter::~FIRFilter()
 // Usual C-version of the filter routine for stereo sound
 uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
 {
-    uint i, j, end;
+    int j, end;
-    LONG_SAMPLETYPE suml, sumr;
 #ifdef SOUNDTOUCH_FLOAT_SAMPLES
     // when using floating point samples, use a scaler instead of a divider
     // because division is much slower operation than multiplying.
@@ -87,9 +86,12 @@ uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, ui
 
     end = 2 * (numSamples - length);
 
+    #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;
@@ -130,28 +132,31 @@ uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, ui
 // Usual C-version of the filter routine for mono sound
 uint FIRFilter::evaluateFilterMono(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
 {
-    uint i, j, end;
+    int j, end;
-    LONG_SAMPLETYPE sum;
 #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);
 
     end = numSamples - length;
+    #pragma omp parallel for
     for (j = 0; j < end; j ++) 
     {
+        const SAMPLETYPE *pSrc = src + j;
+        LONG_SAMPLETYPE sum;
+        uint i;
+
         sum = 0;
         for (i = 0; i < length; i += 4) 
         {
             // loop is unrolled by factor of 4 here for efficiency
-            sum += src[i + 0] * filterCoeffs[i + 0] + 
+            sum += pSrc[i + 0] * filterCoeffs[i + 0] + 
-                   src[i + 1] * filterCoeffs[i + 1] + 
+                   pSrc[i + 1] * filterCoeffs[i + 1] + 
-                   src[i + 2] * filterCoeffs[i + 2] + 
+                   pSrc[i + 2] * filterCoeffs[i + 2] + 
-                   src[i + 3] * filterCoeffs[i + 3];
+                   pSrc[i + 3] * filterCoeffs[i + 3];
         }
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
         sum >>= resultDivFactor;
@@ -161,16 +166,15 @@ uint FIRFilter::evaluateFilterMono(SAMPLETYPE *dest, const SAMPLETYPE *src, uint
         sum *= dScaler;
 #endif // SOUNDTOUCH_INTEGER_SAMPLES
         dest[j] = (SAMPLETYPE)sum;
-        src ++;
     }
     return end;
 }
 
 
-uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) const
+uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels)
 {
-    uint i, j, end, c;
+    int j, end;
-    LONG_SAMPLETYPE *sum=(LONG_SAMPLETYPE*)malloc(numChannels*sizeof(*sum));
+
 #ifdef SOUNDTOUCH_FLOAT_SAMPLES
     // when using floating point samples, use a scaler instead of a divider
     // because division is much slower operation than multiplying.
@@ -181,17 +185,21 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
     assert(src != NULL);
     assert(dest != NULL);
     assert(filterCoeffs != NULL);
+    assert(numChannels < 16);
 
     end = numChannels * (numSamples - length);
 
-    for (c = 0; c < numChannels; c ++)
+    #pragma omp parallel for
-    {
-        sum[c] = 0;
-    }
-
     for (j = 0; j < end; j += numChannels)
     {
         const SAMPLETYPE *ptr;
+        LONG_SAMPLETYPE sums[16];
+        uint c, i;
+
+        for (c = 0; c < numChannels; c ++)
+        {
+            sums[c] = 0;
+        }
 
         ptr = src + j;
 
@@ -200,7 +208,7 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
             SAMPLETYPE coef=filterCoeffs[i];
             for (c = 0; c < numChannels; c ++)
             {
-                sum[c] += ptr[0] * coef;
+                sums[c] += ptr[0] * coef;
                 ptr ++;
             }
         }
@@ -208,13 +216,11 @@ uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uin
         for (c = 0; c < numChannels; c ++)
         {
 #ifdef SOUNDTOUCH_INTEGER_SAMPLES
-            sum[c] >>= resultDivFactor;
+            sums[c] >>= resultDivFactor;
 #else
-            sum[c] *= dScaler;
+            sums[c] *= dScaler;
 #endif // SOUNDTOUCH_INTEGER_SAMPLES
-            *dest = (SAMPLETYPE)sum[c];
+            dest[j+c] = (SAMPLETYPE)sums[c];
-            dest++;
-            sum[c] = 0;
         }
     }
     return numSamples - length;
@@ -253,7 +259,7 @@ uint FIRFilter::getLength() const
 //
 // Note : The amount of outputted samples is by value of 'filter_length' 
 // smaller than the amount of input samples.
-uint FIRFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) const
+uint FIRFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) 
 {
     assert(length > 0);
     assert(lengthDiv8 * 8 == length);

Externals/soundtouch/FIRFilter.h

@@ -11,10 +11,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2013-06-13 01:24:44 +1000 (Thu, 13 Jun 2013) $
+// Last changed  : $Date: 2015-02-21 23:24:29 +0200 (Sat, 21 Feb 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: FIRFilter.h 171 2013-06-12 15:24:44Z oparviai $
+// $Id: FIRFilter.h 202 2015-02-21 21:24:29Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -71,7 +71,7 @@ protected:
     virtual uint evaluateFilterMono(SAMPLETYPE *dest, 
                                     const SAMPLETYPE *src, 
                                     uint numSamples) const;
-    virtual uint evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels) const;
+    virtual uint evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels);
 
 public:
     FIRFilter();
@@ -91,7 +91,7 @@ public:
     uint evaluate(SAMPLETYPE *dest, 
                   const SAMPLETYPE *src, 
                   uint numSamples, 
-                  uint numChannels) const;
+                  uint numChannels);
 
     uint getLength() const;
 

Externals/soundtouch/InterpolateCubic.h

@@ -8,7 +8,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateCubic.h 179 2014-01-06 18:41:42Z oparviai $
+// $Id: InterpolateCubic.h 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -56,7 +56,7 @@ protected:
                         const SAMPLETYPE *src, 
                         int &srcSamples);
 
-    float fract;
+    double fract;
 
 public:
     InterpolateCubic();

Externals/soundtouch/InterpolateLinear.cpp

@@ -8,7 +8,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateLinear.cpp 180 2014-01-06 19:16:02Z oparviai $
+// $Id: InterpolateLinear.cpp 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -170,9 +170,9 @@ int InterpolateLinearInteger::transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE
 
 // Sets new target iRate. Normal iRate = 1.0, smaller values represent slower 
 // iRate, larger faster iRates.
-void InterpolateLinearInteger::setRate(float newRate)
+void InterpolateLinearInteger::setRate(double newRate)
 {
-    iRate = (int)(newRate * SCALE + 0.5f);
+    iRate = (int)(newRate * SCALE + 0.5);
     TransposerBase::setRate(newRate);
 }
 
@@ -190,7 +190,7 @@ InterpolateLinearFloat::InterpolateLinearFloat() : TransposerBase()
     // Notice: use local function calling syntax for sake of clarity, 
     // to indicate the fact that C++ constructor can't call virtual functions.
     resetRegisters();
-    setRate(1.0f);
+    setRate(1.0);
 }
 
 
@@ -275,12 +275,13 @@ int InterpolateLinearFloat::transposeMulti(SAMPLETYPE *dest, const SAMPLETYPE *s
     i = 0;
     while (srcCount < srcSampleEnd)
     {
-        float temp, vol1;
+        float temp, vol1, fract_float;
     
-        vol1 = (1.0f- fract);
+        vol1 = (float)(1.0 - fract);
+		fract_float = (float)fract;
         for (int c = 0; c < numChannels; c ++)
         {
-            temp = vol1 * src[c] + fract * src[c + numChannels];
+			temp = vol1 * src[c] + fract_float * src[c + numChannels];
             *dest = (SAMPLETYPE)temp;
             dest ++;
         }

Externals/soundtouch/InterpolateLinear.h

@@ -8,7 +8,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateLinear.h 179 2014-01-06 18:41:42Z oparviai $
+// $Id: InterpolateLinear.h 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -63,7 +63,7 @@ public:
 
     /// Sets new target rate. Normal rate = 1.0, smaller values represent slower 
     /// rate, larger faster rates.
-    virtual void setRate(float newRate);
+    virtual void setRate(double newRate);
 };
 
 
@@ -71,7 +71,7 @@ public:
 class InterpolateLinearFloat : public TransposerBase
 {
 protected:
-    float fract;
+    double fract;
 
     virtual void resetRegisters();
 

Externals/soundtouch/InterpolateShannon.h

@@ -13,7 +13,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// $Id: InterpolateShannon.h 179 2014-01-06 18:41:42Z oparviai $
+// $Id: InterpolateShannon.h 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -61,7 +61,7 @@ protected:
                         const SAMPLETYPE *src, 
                         int &srcSamples);
 
-    float fract;
+    double fract;
 
 public:
     InterpolateShannon();

Externals/soundtouch/PeakFinder.cpp

@@ -11,10 +11,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2012-12-29 06:52:47 +1100 (Sat, 29 Dec 2012) $
+// Last changed  : $Date: 2015-05-18 18:22:02 +0300 (Mon, 18 May 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: PeakFinder.cpp 164 2012-12-28 19:52:47Z oparviai $
+// $Id: PeakFinder.cpp 213 2015-05-18 15:22:02Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -192,11 +192,21 @@ double PeakFinder::getPeakCenter(const float *data, int peakpos) const
     gp1 = findGround(data, peakpos, -1);
     gp2 = findGround(data, peakpos, 1);
 
-    groundLevel = 0.5f * (data[gp1] + data[gp2]);
     peakLevel = data[peakpos];
 
-    // calculate 70%-level of the peak
+    if (gp1 == gp2) 
-    cutLevel = 0.70f * peakLevel + 0.30f * groundLevel;
+    {
+        // avoid rounding errors when all are equal
+        assert(gp1 == peakpos);
+        cutLevel = groundLevel = peakLevel;
+    } else {
+        // get average of the ground levels
+        groundLevel = 0.5f * (data[gp1] + data[gp2]);
+
+        // calculate 70%-level of the peak
+        cutLevel = 0.70f * peakLevel + 0.30f * groundLevel;
+    }
+
     // find mid-level crossings
     crosspos1 = findCrossingLevel(data, cutLevel, peakpos, -1);
     crosspos2 = findCrossingLevel(data, cutLevel, peakpos, 1);

Externals/soundtouch/PeakFinder.h

@@ -9,7 +9,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2011-12-31 07:33:46 +1100 (Sat, 31 Dec 2011) $
+// 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 $

Externals/soundtouch/RateTransposer.cpp

@@ -10,10 +10,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-04-07 01:57:21 +1000 (Mon, 07 Apr 2014) $
+// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: RateTransposer.cpp 195 2014-04-06 15:57:21Z oparviai $
+// $Id: RateTransposer.cpp 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -97,20 +97,20 @@ AAFilter *RateTransposer::getAAFilter()
 
 // Sets new target iRate. Normal iRate = 1.0, smaller values represent slower 
 // iRate, larger faster iRates.
-void RateTransposer::setRate(float newRate)
+void RateTransposer::setRate(double newRate)
 {
     double fCutoff;
 
     pTransposer->setRate(newRate);
 
     // design a new anti-alias filter
-    if (newRate > 1.0f) 
+    if (newRate > 1.0) 
     {
-        fCutoff = 0.5f / newRate;
+        fCutoff = 0.5 / newRate;
     } 
     else 
     {
-        fCutoff = 0.5f * newRate;
+        fCutoff = 0.5 * newRate;
     }
     pAAFilter->setCutoffFreq(fCutoff);
 }
@@ -225,7 +225,7 @@ void TransposerBase::setAlgorithm(TransposerBase::ALGORITHM a)
 int TransposerBase::transpose(FIFOSampleBuffer &dest, FIFOSampleBuffer &src)
 {
     int numSrcSamples = src.numSamples();
-    int sizeDemand = (int)((float)numSrcSamples / rate) + 8;
+    int sizeDemand = (int)((double)numSrcSamples / rate) + 8;
     int numOutput;
     SAMPLETYPE *psrc = src.ptrBegin();
     SAMPLETYPE *pdest = dest.ptrEnd(sizeDemand);
@@ -270,7 +270,7 @@ void TransposerBase::setChannels(int channels)
 }
 
 
-void TransposerBase::setRate(float newRate)
+void TransposerBase::setRate(double newRate)
 {
     rate = newRate;
 }

Externals/soundtouch/RateTransposer.h

@@ -14,10 +14,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-04-07 01:57:21 +1000 (Mon, 07 Apr 2014) $
+// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: RateTransposer.h 195 2014-04-06 15:57:21Z oparviai $
+// $Id: RateTransposer.h 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -81,14 +81,14 @@ protected:
     static ALGORITHM algorithm;
 
 public:
-    float rate;
+    double rate;
     int numChannels;
 
     TransposerBase();
     virtual ~TransposerBase();
 
     virtual int transpose(FIFOSampleBuffer &dest, FIFOSampleBuffer &src);
-    virtual void setRate(float newRate);
+    virtual void setRate(double newRate);
     virtual void setChannels(int channels);
 
     // static factory function
@@ -158,7 +158,7 @@ public:
 
     /// Sets new target rate. Normal rate = 1.0, smaller values represent slower 
     /// rate, larger faster rates.
-    virtual void setRate(float newRate);
+    virtual void setRate(double newRate);
 
     /// Sets the number of channels, 1 = mono, 2 = stereo
     void setChannels(int channels);

Externals/soundtouch/STTypes.h

@@ -8,10 +8,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2013-06-12 15:24:44 +0000 (Wed, 12 Jun 2013) $
+// Last changed  : $Date: 2015-05-18 18:25:07 +0300 (Mon, 18 May 2015) $
 // File revision : $Revision: 3 $
 //
-// $Id: STTypes.h 171 2013-06-12 15:24:44Z oparviai $
+// $Id: STTypes.h 215 2015-05-18 15:25:07Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -60,16 +60,6 @@ typedef unsigned long   ulong;
     //#include "soundtouch_config.h"
 #endif
 
-#ifndef _WINDEF_
-    // if these aren't defined already by Windows headers, define now
-
-    typedef int BOOL;
-
-    #define FALSE   0
-    #define TRUE    1
-
-#endif  // _WINDEF_
-
 
 namespace soundtouch
 {
@@ -182,6 +172,7 @@ namespace soundtouch
 #else
     // use c++ standard exceptions
     #include <stdexcept>
+    #include <string>
     #define ST_THROW_RT_ERROR(x)    {throw std::runtime_error(x);}
 #endif
 

Externals/soundtouch/SoundTouch.cpp

@@ -41,10 +41,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-04-07 01:57:21 +1000 (Mon, 07 Apr 2014) $
+// Last changed  : $Date: 2015-07-26 17:45:48 +0300 (Sun, 26 Jul 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: SoundTouch.cpp 195 2014-04-06 15:57:21Z oparviai $
+// $Id: SoundTouch.cpp 225 2015-07-26 14:45:48Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -110,6 +110,9 @@ SoundTouch::SoundTouch()
 
     calcEffectiveRateAndTempo();
 
+	samplesExpectedOut = 0;
+	samplesOutput = 0;
+
     channels = 0;
     bSrateSet = false;
 }
@@ -157,7 +160,7 @@ void SoundTouch::setChannels(uint numChannels)
 
 // Sets new rate control value. Normal rate = 1.0, smaller values
 // represent slower rate, larger faster rates.
-void SoundTouch::setRate(float newRate)
+void SoundTouch::setRate(double newRate)
 {
     virtualRate = newRate;
     calcEffectiveRateAndTempo();
@@ -167,9 +170,9 @@ void SoundTouch::setRate(float newRate)
 
 // Sets new rate control value as a difference in percents compared
 // to the original rate (-50 .. +100 %)
-void SoundTouch::setRateChange(float newRate)
+void SoundTouch::setRateChange(double newRate)
 {
-    virtualRate = 1.0f + 0.01f * newRate;
+    virtualRate = 1.0 + 0.01 * newRate;
     calcEffectiveRateAndTempo();
 }
 
@@ -177,7 +180,7 @@ void SoundTouch::setRateChange(float newRate)
 
 // Sets new tempo control value. Normal tempo = 1.0, smaller values
 // represent slower tempo, larger faster tempo.
-void SoundTouch::setTempo(float newTempo)
+void SoundTouch::setTempo(double newTempo)
 {
     virtualTempo = newTempo;
     calcEffectiveRateAndTempo();
@@ -187,9 +190,9 @@ void SoundTouch::setTempo(float newTempo)
 
 // Sets new tempo control value as a difference in percents compared
 // to the original tempo (-50 .. +100 %)
-void SoundTouch::setTempoChange(float newTempo)
+void SoundTouch::setTempoChange(double newTempo)
 {
-    virtualTempo = 1.0f + 0.01f * newTempo;
+    virtualTempo = 1.0 + 0.01 * newTempo;
     calcEffectiveRateAndTempo();
 }
 
@@ -197,7 +200,7 @@ void SoundTouch::setTempoChange(float newTempo)
 
 // Sets new pitch control value. Original pitch = 1.0, smaller values
 // represent lower pitches, larger values higher pitch.
-void SoundTouch::setPitch(float newPitch)
+void SoundTouch::setPitch(double newPitch)
 {
     virtualPitch = newPitch;
     calcEffectiveRateAndTempo();
@@ -207,9 +210,9 @@ void SoundTouch::setPitch(float newPitch)
 
 // Sets pitch change in octaves compared to the original pitch
 // (-1.00 .. +1.00)
-void SoundTouch::setPitchOctaves(float newPitch)
+void SoundTouch::setPitchOctaves(double newPitch)
 {
-    virtualPitch = (float)exp(0.69314718056f * newPitch);
+    virtualPitch = exp(0.69314718056 * newPitch);
     calcEffectiveRateAndTempo();
 }
 
@@ -219,14 +222,14 @@ void SoundTouch::setPitchOctaves(float newPitch)
 // (-12 .. +12)
 void SoundTouch::setPitchSemiTones(int newPitch)
 {
-    setPitchOctaves((float)newPitch / 12.0f);
+    setPitchOctaves((double)newPitch / 12.0);
 }
 
 
 
-void SoundTouch::setPitchSemiTones(float newPitch)
+void SoundTouch::setPitchSemiTones(double newPitch)
 {
-    setPitchOctaves(newPitch / 12.0f);
+    setPitchOctaves(newPitch / 12.0);
 }
 
 
@@ -234,14 +237,14 @@ void SoundTouch::setPitchSemiTones(float newPitch)
 // nominal control values.
 void SoundTouch::calcEffectiveRateAndTempo()
 {
-    float oldTempo = tempo;
+    double oldTempo = tempo;
-    float oldRate = rate;
+    double oldRate = rate;
 
-    tempo = virtualTempo / virtualPitch;
+	tempo = virtualTempo / virtualPitch;
-    rate = virtualPitch * virtualRate;
+	rate = virtualPitch * virtualRate;
 
     if (!TEST_FLOAT_EQUAL(rate,oldRate)) pRateTransposer->setRate(rate);
-    if (!TEST_FLOAT_EQUAL(tempo, oldTempo)) pTDStretch->setTempo(tempo);
+	if (!TEST_FLOAT_EQUAL(tempo, oldTempo)) pTDStretch->setTempo(tempo);
 
 #ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
     if (rate <= 1.0f) 
@@ -317,8 +320,13 @@ void SoundTouch::putSamples(const SAMPLETYPE *samples, uint nSamples)
         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);
+
 #ifndef SOUNDTOUCH_PREVENT_CLICK_AT_RATE_CROSSOVER
-    else if (rate <= 1.0f) 
+    if (rate <= 1.0f) 
     {
         // transpose the rate down, output the transposed sound to tempo changer buffer
         assert(output == pTDStretch);
@@ -346,42 +354,30 @@ void SoundTouch::putSamples(const SAMPLETYPE *samples, uint nSamples)
 void SoundTouch::flush()
 {
     int i;
-    int nUnprocessed;
+	int numStillExpected;
-    int nOut;
+    SAMPLETYPE *buff = new SAMPLETYPE[128 * channels];
-    SAMPLETYPE *buff=(SAMPLETYPE*)malloc(64*channels*sizeof(SAMPLETYPE));
 
-    // check how many samples still await processing, and scale
+	// how many samples are still expected to output
-    // that by tempo & rate to get expected output sample count
+	numStillExpected = (int)((long)(samplesExpectedOut + 0.5) - samplesOutput);
-    nUnprocessed = numUnprocessedSamples();
-    nUnprocessed = (int)((double)nUnprocessed / (tempo * rate) + 0.5);
 
-    nOut = numSamples();        // ready samples currently in buffer ...
+    memset(buff, 0, 128 * channels * sizeof(SAMPLETYPE));
-    nOut += nUnprocessed;       // ... and how many we expect there to be in the end
-    
-    memset(buff, 0, 64 * channels * sizeof(SAMPLETYPE));
     // "Push" the last active samples out from the processing pipeline by
     // feeding blank samples into the processing pipeline until new, 
     // processed samples appear in the output (not however, more than 
-    // 8ksamples in any case)
+    // 24ksamples in any case)
-    for (i = 0; i < 128; i ++) 
+	for (i = 0; (numStillExpected > (int)numSamples()) && (i < 200); i ++)
-    {
+	{
-        putSamples(buff, 64);
+		putSamples(buff, 128);
-        if ((int)numSamples() >= nOut) 
+	}
-        {
-            // Enough new samples have appeared into the output!
-            // As samples come from processing with bigger chunks, now truncate it
-            // back to maximum "nOut" samples to improve duration accuracy 
-            adjustAmountOfSamples(nOut);
 
-            // finish
+	adjustAmountOfSamples(numStillExpected);
-            break;  
-        }
-    }
 
-    // Clear working buffers
+    delete[] buff;
-    pRateTransposer->clear();
+
+    // Clear input buffers
+ //   pRateTransposer->clearInput();
     pTDStretch->clearInput();
-    // yet leave the 'tempoChanger' output intouched as that's where the
+    // yet leave the output intouched as that's where the
     // flushed samples are!
 }
 
@@ -480,6 +476,7 @@ int SoundTouch::getSetting(int settingId) const
 // buffers.
 void SoundTouch::clear()
 {
+	samplesExpectedOut = 0;
     pRateTransposer->clear();
     pTDStretch->clear();
 }
@@ -500,3 +497,30 @@ uint SoundTouch::numUnprocessedSamples() const
     }
     return 0;
 }
+
+
+
+/// 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.
+///
+/// \return Number of samples returned.
+uint SoundTouch::receiveSamples(SAMPLETYPE *output, uint maxSamples)
+{
+	uint ret = FIFOProcessor::receiveSamples(output, maxSamples);
+	samplesOutput += (long)ret;
+	return ret;
+}
+
+
+/// 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.
+uint SoundTouch::receiveSamples(uint maxSamples)
+{
+	uint ret = FIFOProcessor::receiveSamples(maxSamples);
+	samplesOutput += (long)ret;
+	return ret;
+}

Externals/soundtouch/SoundTouch.h

@@ -41,10 +41,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2013-06-12 15:24:44 +0000 (Wed, 12 Jun 2013) $
+// Last changed  : $Date: 2015-09-20 10:38:32 +0300 (Sun, 20 Sep 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: SoundTouch.h 171 2013-06-12 15:24:44Z oparviai $
+// $Id: SoundTouch.h 230 2015-09-20 07:38:32Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -79,10 +79,10 @@ namespace soundtouch
 {
 
 /// Soundtouch library version string
-#define SOUNDTOUCH_VERSION          "1.8.1 (r198)"
+#define SOUNDTOUCH_VERSION          "1.9.2"
 
 /// SoundTouch library version id
-#define SOUNDTOUCH_VERSION_ID       (10801)
+#define SOUNDTOUCH_VERSION_ID       (10902)
 
 //
 // Available setting IDs for the 'setSetting' & 'get_setting' functions:
@@ -151,16 +151,23 @@ private:
     class TDStretch *pTDStretch;
 
     /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters.
-    float virtualRate;
+    double virtualRate;
 
     /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters.
-    float virtualTempo;
+    double virtualTempo;
 
     /// Virtual pitch parameter. Effective rate & tempo are calculated from these parameters.
-    float virtualPitch;
+    double virtualPitch;
 
     /// Flag: Has sample rate been set?
-    BOOL  bSrateSet;
+    bool  bSrateSet;
+
+    /// Accumulator for how many samples in total will be expected as output vs. samples put in,
+    /// considering current processing settings.
+    double samplesExpectedOut;
+
+    /// Accumulator for how many samples in total have been read out from the processing so far
+    long   samplesOutput;
 
     /// Calculates effective rate & tempo valuescfrom 'virtualRate', 'virtualTempo' and 
     /// 'virtualPitch' parameters.
@@ -171,10 +178,10 @@ protected :
     uint  channels;
 
     /// Effective 'rate' value calculated from 'virtualRate', 'virtualTempo' and 'virtualPitch'
-    float rate;
+    double rate;
 
     /// Effective 'tempo' value calculated from 'virtualRate', 'virtualTempo' and 'virtualPitch'
-    float tempo;
+    double tempo;
 
 public:
     SoundTouch();
@@ -188,32 +195,32 @@ public:
 
     /// Sets new rate control value. Normal rate = 1.0, smaller values
     /// represent slower rate, larger faster rates.
-    void setRate(float newRate);
+    void setRate(double newRate);
 
     /// Sets new tempo control value. Normal tempo = 1.0, smaller values
     /// represent slower tempo, larger faster tempo.
-    void setTempo(float newTempo);
+    void setTempo(double newTempo);
 
     /// Sets new rate control value as a difference in percents compared
     /// to the original rate (-50 .. +100 %)
-    void setRateChange(float newRate);
+    void setRateChange(double newRate);
 
     /// Sets new tempo control value as a difference in percents compared
     /// to the original tempo (-50 .. +100 %)
-    void setTempoChange(float newTempo);
+    void setTempoChange(double newTempo);
 
     /// Sets new pitch control value. Original pitch = 1.0, smaller values
     /// represent lower pitches, larger values higher pitch.
-    void setPitch(float newPitch);
+    void setPitch(double newPitch);
 
     /// Sets pitch change in octaves compared to the original pitch  
     /// (-1.00 .. +1.00)
-    void setPitchOctaves(float newPitch);
+    void setPitchOctaves(double newPitch);
 
     /// Sets pitch change in semi-tones compared to the original pitch
     /// (-12 .. +12)
     void setPitchSemiTones(int newPitch);
-    void setPitchSemiTones(float newPitch);
+    void setPitchSemiTones(double newPitch);
 
     /// Sets the number of channels, 1 = mono, 2 = stereo
     void setChannels(uint numChannels);
@@ -240,6 +247,23 @@ public:
                                                     ///< contains data for both channels.
             );
 
+    /// 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.
+    ///
+    /// \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.
+        );
+
+    /// 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.
+        );
+
     /// Clears all the samples in the object's output and internal processing
     /// buffers.
     virtual void clear();
@@ -247,7 +271,7 @@ public:
     /// 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 succesfully changed
     bool setSetting(int settingId,   ///< Setting ID number. see SETTING_... defines.
                     int value        ///< New setting value.
                     );

Externals/soundtouch/TDStretch.cpp

@@ -13,10 +13,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-04-07 01:57:21 +1000 (Mon, 07 Apr 2014) $
+// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
 // File revision : $Revision: 1.12 $
 //
-// $Id: TDStretch.cpp 195 2014-04-06 15:57:21Z oparviai $
+// $Id: TDStretch.cpp 226 2015-08-08 21:00:15Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -63,7 +63,7 @@ using namespace soundtouch;
  *****************************************************************************/
 
 // Table for the hierarchical mixing position seeking algorithm
-static const short _scanOffsets[5][24]={
+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,
@@ -94,7 +94,9 @@ TDStretch::TDStretch() : FIFOProcessor(&outputBuffer)
     bAutoSeqSetting = true;
     bAutoSeekSetting = true;
 
-//    outDebt = 0;
+    maxnorm = 0;
+    maxnormf = 1e8;
+
     skipFract = 0;
 
     tempo = 1.0f;
@@ -250,7 +252,7 @@ int TDStretch::seekBestOverlapPosition(const SAMPLETYPE *refPos)
     if (bQuickSeek) 
     {
         return seekBestOverlapPositionQuick(refPos);
-    } 
+    }
     else 
     {
         return seekBestOverlapPositionFull(refPos);
@@ -282,7 +284,6 @@ inline void TDStretch::overlap(SAMPLETYPE *pOutput, const SAMPLETYPE *pInput, ui
 }
 
 
-
 // Seeks for the optimal overlap-mixing position. The 'stereo' version of the
 // routine
 //
@@ -292,9 +293,9 @@ inline void TDStretch::overlap(SAMPLETYPE *pOutput, const SAMPLETYPE *pInput, ui
 int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos) 
 {
     int bestOffs;
-    double bestCorr, corr;
+    double bestCorr;
-    double norm;
     int i;
+    double norm;
 
     bestCorr = FLT_MIN;
     bestOffs = 0;
@@ -302,14 +303,22 @@ int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos)
     // Scans for the best correlation value by testing each possible position
     // over the permitted range.
     bestCorr = calcCrossCorr(refPos, pMidBuffer, norm);
+
+    #pragma omp parallel for
     for (i = 1; i < seekLength; i ++) 
     {
-        // Calculates correlation value for the mixing position corresponding
+        double corr;
-        // to 'i'. Now call "calcCrossCorrAccumulate" that is otherwise same as
+        // Calculates correlation value for the mixing position corresponding to 'i'
-        // "calcCrossCorr", but saves time by reusing & updating previously stored 
+#ifdef _OPENMP
+        // in parallel OpenMP mode, can't use norm accumulator version as parallel executor won't
+        // iterate the loop in sequential order
+        corr = calcCrossCorr(refPos + channels * i, pMidBuffer, norm);
+#else
+        // In non-parallel version call "calcCrossCorrAccumulate" that is otherwise same
+        // as "calcCrossCorr", but saves time by reusing & updating previously stored 
         // "norm" value
         corr = calcCrossCorrAccumulate(refPos + channels * i, pMidBuffer, norm);
-
+#endif
         // heuristic rule to slightly favour values close to mid of the range
         double tmp = (double)(2 * i - seekLength) / (double)seekLength;
         corr = ((corr + 0.1) * (1.0 - 0.25 * tmp * tmp));
@@ -317,75 +326,184 @@ int TDStretch::seekBestOverlapPositionFull(const SAMPLETYPE *refPos)
         // Checks for the highest correlation value
         if (corr > bestCorr) 
         {
+            // For optimal performance, enter critical section only in case that best value found.
+            // in such case repeat 'if' condition as it's possible that parallel execution may have
+            // updated the bestCorr value in the mean time
+            #pragma omp critical
+            if (corr > bestCorr)
+            {
+                bestCorr = corr;
+                bestOffs = i;
+            }
+        }
+    }
+
+#ifdef SOUNDTOUCH_INTEGER_SAMPLES
+    adaptNormalizer();
+#endif
+
+    // clear cross correlation routine state if necessary (is so e.g. in MMX routines).
+    clearCrossCorrState();
+
+    return bestOffs;
+}
+
+
+// Quick seek algorithm for improved runtime-performance: First roughly scans through the 
+// correlation area, and then scan surroundings of two best preliminary correlation candidates
+// with improved precision
+//
+// Based on testing:
+// - This algorithm gives on average 99% as good match as the full algorith
+// - 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
+int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos)
+{
+#define _MIN(a, b)   (((a) < (b)) ? (a) : (b))
+#define SCANSTEP    16
+#define SCANWIND    8
+
+    int bestOffs;
+    int i;
+    int bestOffs2;
+    float bestCorr, corr;
+    float bestCorr2;
+    double norm;
+
+    // note: 'float' types used in this function in case that the platform would need to use software-fp
+
+    bestCorr = FLT_MIN;
+    bestOffs = SCANWIND;
+    bestCorr2 = FLT_MIN;
+    bestOffs2 = 0;
+
+    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
+    // increase possibility of ideal match.
+    //
+    // Begin from "SCANSTEP" instead of SCANWIND to make the calculation
+    // catch the 'middlepoint' of seekLength vector as that's the a-priori 
+    // expected best match position
+    //
+    // Roughly:
+    // - 15% of cases find best result directly on the first round,
+    // - 75% cases find better match on 2nd round around the best match from 1st round
+    // - 10% cases find better match on 2nd round around the 2nd-best-match from 1st round
+    for (i = SCANSTEP; i < seekLength - SCANWIND - 1; i += SCANSTEP)
+    {
+        // Calculates correlation value for the mixing position corresponding
+        // to 'i'
+        corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm);
+        // heuristic rule to slightly favour values close to mid of the seek range
+        float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength;
+        corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp));
+
+        // Checks for the highest correlation value
+        if (corr > bestCorr)
+        {
+            // found new best match. keep the previous best as 2nd best match
+            bestCorr2 = bestCorr;
+            bestOffs2 = bestOffs;
             bestCorr = corr;
             bestOffs = i;
         }
-    }
+        else if (corr > bestCorr2)
-    // clear cross correlation routine state if necessary (is so e.g. in MMX routines).
-    clearCrossCorrState();
-
-    return bestOffs;
-}
-
-
-// Seeks for the optimal overlap-mixing position. The 'stereo' version of the
-// routine
-//
-// The best position is determined as the position where the two overlapped
-// sample sequences are 'most alike', in terms of the highest cross-correlation
-// value over the overlapping period
-int TDStretch::seekBestOverlapPositionQuick(const SAMPLETYPE *refPos) 
-{
-    int j;
-    int bestOffs;
-    double bestCorr, corr;
-    int scanCount, corrOffset, tempOffset;
-
-    bestCorr = FLT_MIN;
-    bestOffs = _scanOffsets[0][0];
-    corrOffset = 0;
-    tempOffset = 0;
-
-    // Scans for the best correlation value using four-pass hierarchical search.
-    //
-    // The look-up table 'scans' has hierarchical position adjusting steps.
-    // In first pass the routine searhes for the highest correlation with 
-    // relatively coarse steps, then rescans the neighbourhood of the highest
-    // correlation with better resolution and so on.
-    for (scanCount = 0;scanCount < 4; scanCount ++) 
-    {
-        j = 0;
-        while (_scanOffsets[scanCount][j]) 
         {
-            double norm;
+            // not new best, but still new 2nd best match
-            tempOffset = corrOffset + _scanOffsets[scanCount][j];
+            bestCorr2 = corr;
-            if (tempOffset >= seekLength) break;
+            bestOffs2 = i;
-
-            // Calculates correlation value for the mixing position corresponding
-            // to 'tempOffset'
-            corr = (double)calcCrossCorr(refPos + channels * tempOffset, pMidBuffer, norm);
-            // heuristic rule to slightly favour values close to mid of the range
-            double tmp = (double)(2 * tempOffset - seekLength) / seekLength;
-            corr = ((corr + 0.1) * (1.0 - 0.25 * tmp * tmp));
-
-            // Checks for the highest correlation value
-            if (corr > bestCorr) 
-            {
-                bestCorr = corr;
-                bestOffs = tempOffset;
-            }
-            j ++;
         }
-        corrOffset = bestOffs;
     }
+
+    // Scans surroundings of the found best match with small stepping
+    int end = _MIN(bestOffs + SCANWIND + 1, seekLength);
+    for (i = bestOffs - SCANWIND; i < end; i++)
+    {
+        if (i == bestOffs) continue;    // this offset already calculated, thus skip
+
+        // Calculates correlation value for the mixing position corresponding
+        // to 'i'
+        corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm);
+        // heuristic rule to slightly favour values close to mid of the range
+        float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength;
+        corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp));
+
+        // Checks for the highest correlation value
+        if (corr > bestCorr)
+        {
+            bestCorr = corr;
+            bestOffs = i;
+            best = 1;
+        }
+    }
+
+    // Scans surroundings of the 2nd best match with small stepping
+    end = _MIN(bestOffs2 + SCANWIND + 1, seekLength);
+    for (i = bestOffs2 - SCANWIND; i < end; i++)
+    {
+        if (i == bestOffs2) continue;    // this offset already calculated, thus skip
+
+        // Calculates correlation value for the mixing position corresponding
+        // to 'i'
+        corr = (float)calcCrossCorr(refPos + channels*i, pMidBuffer, norm);
+        // heuristic rule to slightly favour values close to mid of the range
+        float tmp = (float)(2 * i - seekLength - 1) / (float)seekLength;
+        corr = ((corr + 0.1f) * (1.0f - 0.25f * tmp * tmp));
+
+        // Checks for the highest correlation value
+        if (corr > bestCorr)
+        {
+            bestCorr = corr;
+            bestOffs = i;
+            best = 2;
+        }
+    }
+
     // clear cross correlation routine state if necessary (is so e.g. in MMX routines).
     clearCrossCorrState();
 
+#ifdef SOUNDTOUCH_INTEGER_SAMPLES
+    adaptNormalizer();
+#endif
+
     return bestOffs;
 }
 
 
 
+
+/// For integer algorithm: adapt normalization factor divider with music so that 
+/// it'll not be pessimistically restrictive that can degrade quality on quieter sections
+/// yet won't cause integer overflows either
+void TDStretch::adaptNormalizer()
+{
+    // Do not adapt normalizer over too silent sequences to avoid averaging filter depleting to
+    // too low values during pauses in music
+    if ((maxnorm > 1000) || (maxnormf > 40000000))
+    { 
+        //norm averaging filter
+        maxnormf = 0.9f * maxnormf + 0.1f * (float)maxnorm;
+
+        if ((maxnorm > 800000000) && (overlapDividerBitsNorm < 16))
+        {
+            // large values, so increase divider
+            overlapDividerBitsNorm++;
+            if (maxnorm > 1600000000) overlapDividerBitsNorm++; // extra large value => extra increase
+        }
+        else if ((maxnormf < 1000000) && (overlapDividerBitsNorm > 0))
+        {
+            // extra small values, decrease divider
+            overlapDividerBitsNorm--;
+        }
+    }
+
+    maxnorm = 0;
+}
+
+
 /// clear cross correlation routine state if necessary 
 void TDStretch::clearCrossCorrState()
 {
@@ -407,7 +525,7 @@ void TDStretch::calcSeqParameters()
     #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 tempo
+    // seek-window-ms setting values at above low & top tempoq
     #define AUTOSEEK_AT_MIN     25.0
     #define AUTOSEEK_AT_MAX     15.0
     #define AUTOSEEK_K          ((AUTOSEEK_AT_MAX - AUTOSEEK_AT_MIN) / (AUTOSEQ_TEMPO_TOP - AUTOSEQ_TEMPO_LOW))
@@ -444,7 +562,7 @@ void TDStretch::calcSeqParameters()
 
 // Sets new target tempo. Normal tempo = 'SCALE', smaller values represent slower 
 // tempo, larger faster tempo.
-void TDStretch::setTempo(float newTempo)
+void TDStretch::setTempo(double newTempo)
 {
     int intskip;
 
@@ -455,7 +573,7 @@ void TDStretch::setTempo(float newTempo)
 
     // Calculate ideal skip length (according to tempo value) 
     nominalSkip = tempo * (seekWindowLength - overlapLength);
-    intskip = (int)(nominalSkip + 0.5f);
+    intskip = (int)(nominalSkip + 0.5);
 
     // Calculate how many samples are needed in the 'inputBuffer' to 
     // process another batch of samples
@@ -721,13 +839,15 @@ void TDStretch::calculateOverlapLength(int aoverlapMs)
     // calculate overlap length so that it's power of 2 - thus it's easy to do
     // integer division by right-shifting. Term "-1" at end is to account for 
     // the extra most significatnt bit left unused in result by signed multiplication 
-    overlapDividerBits = _getClosest2Power((sampleRate * aoverlapMs) / 1000.0) - 1;
+    overlapDividerBitsPure = _getClosest2Power((sampleRate * aoverlapMs) / 1000.0) - 1;
-    if (overlapDividerBits > 9) overlapDividerBits = 9;
+    if (overlapDividerBitsPure > 9) overlapDividerBitsPure = 9;
-    if (overlapDividerBits < 3) overlapDividerBits = 3;
+    if (overlapDividerBitsPure < 3) overlapDividerBitsPure = 3;
-    newOvl = (int)pow(2.0, (int)overlapDividerBits + 1);    // +1 => account for -1 above
+    newOvl = (int)pow(2.0, (int)overlapDividerBitsPure + 1);    // +1 => account for -1 above
 
     acceptNewOverlapLength(newOvl);
 
+    overlapDividerBitsNorm = overlapDividerBitsPure;
+
     // calculate sloping divider so that crosscorrelation operation won't 
     // overflow 32-bit register. Max. sum of the crosscorrelation sum without 
     // divider would be 2^30*(N^3-N)/3, where N = overlap length
@@ -735,10 +855,10 @@ void TDStretch::calculateOverlapLength(int aoverlapMs)
 }
 
 
-double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, double &norm) const
+double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, double &norm)
 {
     long corr;
-    long lnorm;
+    unsigned long lnorm;
     int i;
 
     corr = lnorm = 0;
@@ -748,15 +868,19 @@ double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, do
     for (i = 0; i < channels * overlapLength; i += 4) 
     {
         corr += (mixingPos[i] * compare[i] + 
-                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBits;  // notice: do intermediate division here to avoid integer overflow
+                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;  // notice: do intermediate division here to avoid integer overflow
         corr += (mixingPos[i + 2] * compare[i + 2] + 
-                 mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBits;
+                mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm;
         lnorm += (mixingPos[i] * mixingPos[i] + 
-                  mixingPos[i + 1] * mixingPos[i + 1]) >> overlapDividerBits; // notice: do intermediate division here to avoid integer overflow
+                mixingPos[i + 1] * mixingPos[i + 1]) >> overlapDividerBitsNorm; // notice: do intermediate division here to avoid integer overflow
         lnorm += (mixingPos[i + 2] * mixingPos[i + 2] + 
-                  mixingPos[i + 3] * mixingPos[i + 3]) >> overlapDividerBits;
+                mixingPos[i + 3] * mixingPos[i + 3]) >> overlapDividerBitsNorm;
     }
 
+    if (lnorm > maxnorm)
+    {
+        maxnorm = lnorm;
+    }
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
     norm = (double)lnorm;
@@ -765,17 +889,17 @@ double TDStretch::calcCrossCorr(const short *mixingPos, const short *compare, do
 
 
 /// Update cross-correlation by accumulating "norm" coefficient by previously calculated value
-double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm) const
+double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm)
 {
     long corr;
-    long lnorm;
+    unsigned long lnorm;
     int i;
 
     // cancel first normalizer tap from previous round
     lnorm = 0;
     for (i = 1; i <= channels; i ++)
     {
-        lnorm -= (mixingPos[-i] * mixingPos[-i]) >> overlapDividerBits;
+        lnorm -= (mixingPos[-i] * mixingPos[-i]) >> overlapDividerBitsNorm;
     }
 
     corr = 0;
@@ -785,18 +909,23 @@ double TDStretch::calcCrossCorrAccumulate(const short *mixingPos, const short *c
     for (i = 0; i < channels * overlapLength; i += 4) 
     {
         corr += (mixingPos[i] * compare[i] + 
-                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBits;  // notice: do intermediate division here to avoid integer overflow
+                 mixingPos[i + 1] * compare[i + 1]) >> overlapDividerBitsNorm;  // notice: do intermediate division here to avoid integer overflow
         corr += (mixingPos[i + 2] * compare[i + 2] + 
-                 mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBits;
+                 mixingPos[i + 3] * compare[i + 3]) >> overlapDividerBitsNorm;
     }
 
     // update normalizer with last samples of this round
     for (int j = 0; j < channels; j ++)
     {
         i --;
-        lnorm += (mixingPos[i] * mixingPos[i]) >> overlapDividerBits;
+        lnorm += (mixingPos[i] * mixingPos[i]) >> overlapDividerBitsNorm;
     }
+
     norm += (double)lnorm;
+    if (norm > maxnorm)
+    {
+        maxnorm = (unsigned long)norm;
+    }
 
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
@@ -881,9 +1010,10 @@ void TDStretch::calculateOverlapLength(int overlapInMsec)
 
 
 /// Calculate cross-correlation
-double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, double &norm) const
+double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, double &anorm)
 {
     double corr;
+    double norm;
     int i;
 
     corr = norm = 0;
@@ -905,12 +1035,13 @@ double TDStretch::calcCrossCorr(const float *mixingPos, const float *compare, do
                 mixingPos[i + 3] * mixingPos[i + 3];
     }
 
+    anorm = norm;
     return corr / sqrt((norm < 1e-9 ? 1.0 : norm));
 }
 
 
 /// Update cross-correlation by accumulating "norm" coefficient by previously calculated value
-double TDStretch::calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm) const
+double TDStretch::calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm)
 {
     double corr;
     int i;

Externals/soundtouch/TDStretch.h

@@ -13,10 +13,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-04-07 01:57:21 +1000 (Mon, 07 Apr 2014) $
+// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: TDStretch.h 195 2014-04-06 15:57:21Z oparviai $
+// $Id: TDStretch.h 226 2015-08-08 21:00:15Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -112,39 +112,46 @@ class TDStretch : public FIFOProcessor
 protected:
     int channels;
     int sampleReq;
-    float tempo;
 
-    SAMPLETYPE *pMidBuffer;
-    SAMPLETYPE *pMidBufferUnaligned;
     int overlapLength;
     int seekLength;
     int seekWindowLength;
-    int overlapDividerBits;
+    int overlapDividerBitsNorm;
+    int overlapDividerBitsPure;
     int slopingDivider;
-    float nominalSkip;
-    float skipFract;
-    FIFOSampleBuffer outputBuffer;
-    FIFOSampleBuffer inputBuffer;
-    bool bQuickSeek;
-
     int sampleRate;
     int sequenceMs;
     int seekWindowMs;
     int overlapMs;
+
+    unsigned long maxnorm;
+    float maxnormf;
+
+    double tempo;
+    double nominalSkip;
+    double skipFract;
+
+    bool bQuickSeek;
     bool bAutoSeqSetting;
     bool bAutoSeekSetting;
 
+    SAMPLETYPE *pMidBuffer;
+    SAMPLETYPE *pMidBufferUnaligned;
+
+    FIFOSampleBuffer outputBuffer;
+    FIFOSampleBuffer inputBuffer;
+
     void acceptNewOverlapLength(int newOverlapLength);
 
     virtual void clearCrossCorrState();
     void calculateOverlapLength(int overlapMs);
 
-    virtual double calcCrossCorr(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm) const;
+    virtual double calcCrossCorr(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm);
-    virtual double calcCrossCorrAccumulate(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm) const;
+    virtual double calcCrossCorrAccumulate(const SAMPLETYPE *mixingPos, const SAMPLETYPE *compare, double &norm);
 
     virtual int seekBestOverlapPositionFull(const SAMPLETYPE *refPos);
     virtual int seekBestOverlapPositionQuick(const SAMPLETYPE *refPos);
-    int seekBestOverlapPosition(const SAMPLETYPE *refPos);
+    virtual int seekBestOverlapPosition(const SAMPLETYPE *refPos);
 
     virtual void overlapStereo(SAMPLETYPE *output, const SAMPLETYPE *input) const;
     virtual void overlapMono(SAMPLETYPE *output, const SAMPLETYPE *input) const;
@@ -154,6 +161,8 @@ protected:
     void overlap(SAMPLETYPE *output, const SAMPLETYPE *input, uint ovlPos) const;
 
     void calcSeqParameters();
+    void adaptNormalizer();
+
 
     /// Changes the tempo of the given sound samples.
     /// Returns amount of samples returned in the "output" buffer.
@@ -182,7 +191,7 @@ public:
 
     /// Sets new target tempo. Normal tempo = 'SCALE', smaller values represent slower 
     /// tempo, larger faster tempo.
-    void setTempo(float newTempo);
+    void setTempo(double newTempo);
 
     /// Returns nonzero if there aren't any samples available for outputting.
     virtual void clear();
@@ -249,8 +258,8 @@ public:
     class TDStretchMMX : public TDStretch
     {
     protected:
-        double calcCrossCorr(const short *mixingPos, const short *compare, double &norm) const;
+        double calcCrossCorr(const short *mixingPos, const short *compare, double &norm);
-        double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm) const;
+        double calcCrossCorrAccumulate(const short *mixingPos, const short *compare, double &norm);
         virtual void overlapStereo(short *output, const short *input) const;
         virtual void clearCrossCorrState();
     };
@@ -262,8 +271,8 @@ public:
     class TDStretchSSE : public TDStretch
     {
     protected:
-        double calcCrossCorr(const float *mixingPos, const float *compare, double &norm) const;
+        double calcCrossCorr(const float *mixingPos, const float *compare, double &norm);
-        double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm) const;
+        double calcCrossCorrAccumulate(const float *mixingPos, const float *compare, double &norm);
     };
 
 #endif /// SOUNDTOUCH_ALLOW_SSE

Externals/soundtouch/cpu_detect.h

@@ -12,7 +12,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2008-02-11 03:26:55 +1100 (Mon, 11 Feb 2008) $
+// 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 $

Externals/soundtouch/cpu_detect_x86.cpp

@@ -11,7 +11,7 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-08 05:24:28 +1100 (Wed, 08 Jan 2014) $
+// 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 $

Externals/soundtouch/mmx_optimized.cpp

@@ -20,10 +20,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-08 05:25:40 +1100 (Wed, 08 Jan 2014) $
+// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: mmx_optimized.cpp 184 2014-01-07 18:25:40Z oparviai $
+// $Id: mmx_optimized.cpp 226 2015-08-08 21:00:15Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -68,7 +68,7 @@ using namespace soundtouch;
 
 
 // Calculates cross correlation of two buffers
-double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &dnorm) const
+double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &dnorm)
 {
     const __m64 *pVec1, *pVec2;
     __m64 shifter;
@@ -79,7 +79,7 @@ double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &d
     pVec1 = (__m64*)pV1;
     pVec2 = (__m64*)pV2;
 
-    shifter = _m_from_int(overlapDividerBits);
+    shifter = _m_from_int(overlapDividerBitsNorm);
     normaccu = accu = _mm_setzero_si64();
 
     // Process 4 parallel sets of 2 * stereo samples or 4 * mono samples 
@@ -123,6 +123,11 @@ double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &d
     // Clear MMS state
     _m_empty();
 
+    if (norm > (long)maxnorm)
+    {
+        maxnorm = norm;
+    }
+
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
     dnorm = (double)norm;
@@ -134,7 +139,7 @@ double TDStretchMMX::calcCrossCorr(const short *pV1, const short *pV2, double &d
 
 
 /// Update cross-correlation by accumulating "norm" coefficient by previously calculated value
-double TDStretchMMX::calcCrossCorrAccumulate(const short *pV1, const short *pV2, double &dnorm) const
+double TDStretchMMX::calcCrossCorrAccumulate(const short *pV1, const short *pV2, double &dnorm)
 {
     const __m64 *pVec1, *pVec2;
     __m64 shifter;
@@ -146,13 +151,13 @@ double TDStretchMMX::calcCrossCorrAccumulate(const short *pV1, const short *pV2,
     lnorm = 0;
     for (i = 1; i <= channels; i ++)
     {
-        lnorm -= (pV1[-i] * pV1[-i]) >> overlapDividerBits;
+        lnorm -= (pV1[-i] * pV1[-i]) >> overlapDividerBitsNorm;
     }
 
     pVec1 = (__m64*)pV1;
     pVec2 = (__m64*)pV2;
 
-    shifter = _m_from_int(overlapDividerBits);
+    shifter = _m_from_int(overlapDividerBitsNorm);
     accu = _mm_setzero_si64();
 
     // Process 4 parallel sets of 2 * stereo samples or 4 * mono samples 
@@ -191,10 +196,15 @@ double TDStretchMMX::calcCrossCorrAccumulate(const short *pV1, const short *pV2,
     pV1 = (short *)pVec1;
     for (int j = 1; j <= channels; j ++)
     {
-        lnorm += (pV1[-j] * pV1[-j]) >> overlapDividerBits;
+        lnorm += (pV1[-j] * pV1[-j]) >> overlapDividerBitsNorm;
     }
     dnorm += (double)lnorm;
 
+    if (lnorm > (long)maxnorm)
+    {
+        maxnorm = lnorm;
+    }
+
     // Normalize result by dividing by sqrt(norm) - this step is easiest 
     // done using floating point operation
     return (double)corr / sqrt((dnorm < 1e-9) ? 1.0 : dnorm);
@@ -233,7 +243,7 @@ void TDStretchMMX::overlapStereo(short *output, const short *input) const
 
     // Overlaplength-division by shifter. "+1" is to account for "-1" deduced in
     // overlapDividerBits calculation earlier.
-    shifter = _m_from_int(overlapDividerBits + 1);
+    shifter = _m_from_int(overlapDividerBitsPure + 1);
 
     for (i = 0; i < overlapLength / 4; i ++)
     {
@@ -287,6 +297,7 @@ void TDStretchMMX::overlapStereo(short *output, const short *input) const
 
 FIRFilterMMX::FIRFilterMMX() : FIRFilter()
 {
+    filterCoeffsAlign = NULL;
     filterCoeffsUnalign = NULL;
 }
 

Externals/soundtouch/sse_optimized.cpp

@@ -23,10 +23,10 @@
 ///
 ////////////////////////////////////////////////////////////////////////////////
 //
-// Last changed  : $Date: 2014-01-08 05:25:40 +1100 (Wed, 08 Jan 2014) $
+// Last changed  : $Date: 2015-08-09 00:00:15 +0300 (Sun, 09 Aug 2015) $
 // File revision : $Revision: 4 $
 //
-// $Id: sse_optimized.cpp 184 2014-01-07 18:25:40Z oparviai $
+// $Id: sse_optimized.cpp 226 2015-08-08 21:00:15Z oparviai $
 //
 ////////////////////////////////////////////////////////////////////////////////
 //
@@ -71,7 +71,7 @@ using namespace soundtouch;
 #include <math.h>
 
 // Calculates cross correlation of two buffers
-double TDStretchSSE::calcCrossCorr(const float *pV1, const float *pV2, double &norm) const
+double TDStretchSSE::calcCrossCorr(const float *pV1, const float *pV2, double &anorm)
 {
     int i;
     const float *pVec1;
@@ -141,7 +141,8 @@ double TDStretchSSE::calcCrossCorr(const float *pV1, const float *pV2, double &n
 
     // return value = vSum[0] + vSum[1] + vSum[2] + vSum[3]
     float *pvNorm = (float*)&vNorm;
-    norm = (pvNorm[0] + pvNorm[1] + pvNorm[2] + pvNorm[3]);
+    float norm = (pvNorm[0] + pvNorm[1] + pvNorm[2] + pvNorm[3]);
+    anorm = norm;
 
     float *pvSum = (float*)&vSum;
     return (double)(pvSum[0] + pvSum[1] + pvSum[2] + pvSum[3]) / sqrt(norm < 1e-9 ? 1.0 : norm);
@@ -182,7 +183,7 @@ double TDStretchSSE::calcCrossCorr(const float *pV1, const float *pV2, double &n
 
 
 
-double TDStretchSSE::calcCrossCorrAccumulate(const float *pV1, const float *pV2, double &norm) const
+double TDStretchSSE::calcCrossCorrAccumulate(const float *pV1, const float *pV2, double &norm)
 {
     // call usual calcCrossCorr function because SSE does not show big benefit of 
     // accumulating "norm" value, and also the "norm" rolling algorithm would get 
@@ -258,14 +259,17 @@ uint FIRFilterSSE::evaluateFilterStereo(float *dest, const float *source, uint n
     assert(((ulongptr)filterCoeffsAlign) % 16 == 0);
 
     // filter is evaluated for two stereo samples with each iteration, thus use of 'j += 2'
+    #pragma omp parallel for
     for (j = 0; j < count; j += 2)
     {
         const float *pSrc;
+        float *pDest;
         const __m128 *pFil;
         __m128 sum1, sum2;
         uint i;
 
-        pSrc = (const float*)source;              // source audio data
+        pSrc = (const float*)source + j * 2;      // source audio data
+        pDest = dest + j * 2;                     // destination audio data
         pFil = (const __m128*)filterCoeffsAlign;  // filter coefficients. NOTE: Assumes coefficients 
                                                   // are aligned to 16-byte boundary
         sum1 = sum2 = _mm_setzero_ps();
@@ -298,12 +302,10 @@ uint FIRFilterSSE::evaluateFilterStereo(float *dest, const float *source, uint n
         // to sum the two hi- and lo-floats of these registers together.
 
         // post-shuffle & add the filtered values and store to dest.
-        _mm_storeu_ps(dest, _mm_add_ps(
+        _mm_storeu_ps(pDest, _mm_add_ps(
                     _mm_shuffle_ps(sum1, sum2, _MM_SHUFFLE(1,0,3,2)),   // s2_1 s2_0 s1_3 s1_2
                     _mm_shuffle_ps(sum1, sum2, _MM_SHUFFLE(3,2,1,0))    // s2_3 s2_2 s1_1 s1_0
                     ));
-        source += 4;
-        dest += 4;
     }
 
     // Ideas for further improvement:

Source/Core/AudioCommon/OpenALStream.cpp

@@ -233,11 +233,11 @@ void OpenALStream::SoundLoop()
 
 		if (iBuffersProcessed)
 		{
-			float rate = m_mixer->GetCurrentSpeed();
+			double rate = (double)m_mixer->GetCurrentSpeed();
 			if (rate <= 0)
 			{
 				Core::RequestRefreshInfo();
-				rate = m_mixer->GetCurrentSpeed();
+				rate = (double)m_mixer->GetCurrentSpeed();
 			}
 
 			// Place a lower limit of 10% speed.  When a game boots up, there will be