diff --git a/README.md b/README.md
index 6e61997..25966f9 100644
--- a/README.md
+++ b/README.md
@@ -34,7 +34,7 @@ exhale is being made available under an open-source license which is
similar to the 3-clause BSD license but modified to address specific
aspects dictated by the nature and the output of this application.
-The license text and release notes for the current version 1.0.2 can
+The license text and release notes for the current version 1.0.3 can
be found in the `include` subdirectory of the exhale distribution.
diff --git a/include/Release.htm b/include/Release.htm
index 9663a96..6f41287 100644
--- a/include/Release.htm
+++ b/include/Release.htm
@@ -25,9 +25,13 @@
exhale - ecodis extended high-efficiency and low-complexity encoder
Software Release Notes, Version History, Known Issues, Upcoming Feature Roadmap
- The version of this distribution of the «exhale» software release is 1.0.2 (official public minor release) from March 2020. Please check www.ecodis.de regularly for new versions of this software. A summary of each version up to this release, a list of known issues with this release, and a roadmap of additional functionality are provided below.
+ The version of this distribution of the «exhale» software release is 1.0.3 (official public minor release) from April 2020. Please check www.ecodis.de regularly for new versions of this software. A summary of each version up to this release, a list of known issues with this release, and a roadmap of additional functionality are provided below.
Chronological Version History
- Version 1.0.2 Mar. 2020, this release
+ Version 1.0.3 Apr. 2020, this release
+
+ extended basic joint-stereo coding functionality for mid/high rates, minor bugfixes
+ Version 1.0.2 Mar. 2020
added basic low/mid-rate joint-stereo coding functionality, bugfixes, and speedupsexhaleApp: support for input sampling rates of up to 48000 Hz with CVBR mode 2 If you are in need of an additional library or application feature not mentioned below, please contact ecodis or a contributor with a request, and we will see what we can do.
support for coding with a core coder frame length of 768 samples, no version planexhaleLib: finalized integration of higher-rate joint-channel coding, version 1.0.3exhaleLib: quality tuning and bug fixing for low-rate stereo coding, version 1.0.3exhaleLib: finalization of support for 3.0 – 5.1 multichannel coding, version 1.0.4exhaleLib: completed integration of predictive joint-channel coding, version 1.0.4exhaleLib: quality tuning and bug fixing for low-rate stereo coding, version 1.0.4exhaleLib: finalization of support for 3.0 – 5.1 multichannel coding, version 1.0.5exhaleLib: speed-ups and further quality tuning for critical signals, version 1.0.5.
-Written by C. R. Helmrich for exhale 1.0.2, Mar. 2020. Available at www.ecodis.de/exhale/release.htm.
+Written by C. R. Helmrich for exhale 1.0.3, Apr. 2020. Available at www.ecodis.de/exhale/release.htm.
|
diff --git a/src/lib/exhaleEnc.cpp b/src/lib/exhaleEnc.cpp
index 7586499..b24d3ad 100644
--- a/src/lib/exhaleEnc.cpp
+++ b/src/lib/exhaleEnc.cpp
@@ -693,7 +693,7 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
const unsigned lfeChannelIndex = (m_channelConf >= CCI_6_CH ? __max (5, nChannels - 1) : USAC_MAX_NUM_CHANNELS);
const uint32_t maxSfbLong = (samplingRate < 37566 ? 51 /*32 kHz*/ : brModeAndFsToMaxSfbLong (m_bitRateMode, samplingRate));
const uint32_t reductionFactor = (samplingRate < 37566 ? 2 : 3); // undercoding reduction
- const uint64_t scaleSr = (samplingRate < 27713 ? 37 - m_bitRateMode : 37) - (m_bitRateMode > 3 ? nChannels >> 1 : 0);
+ const uint64_t scaleSr = (samplingRate < 27713 ? 37 - m_bitRateMode : 37) - (m_bitRateMode > 2 ? nChannels >> 1 : 0);
const uint64_t scaleBr = (m_bitRateMode == 0 ? 32 : scaleSr - eightTimesSqrt256Minus[256 - m_bitRateMode] - __min (3, (m_bitRateMode - 1) >> 1));
uint32_t* sfbStepSizes = (uint32_t*) m_tempIntBuf;
uint8_t meanSpecFlat[USAC_MAX_NUM_CHANNELS];
@@ -857,7 +857,9 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
const uint32_t* refRms = &coreConfig.groupingData[1 - ch].sfbRmsValues[m_numSwbShort * gr];
uint8_t* grpScaleFacs = &grpData.scaleFactors[m_numSwbShort * gr];
uint32_t* grpStepSizes = &stepSizes[m_numSwbShort * gr];
+#if SA_IMPROVED_REAL_ABS
uint8_t* grpStereoData = &coreConfig.stereoDataCurr[m_numSwbShort * gr];
+#endif
uint32_t b, grpRmsMin = INT_MAX; // min. RMS value, used for overcoding reduction
// undercoding reduction for case where large number of coefs is quantized to zero
@@ -873,6 +875,9 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
getComplexRmsValue (refRms[b], gr, b, numSwbCh, coreConfig.tnsData[1 - ch]) >> 9);
#endif
if (grpRms[b] < grpRmsMin) grpRmsMin = grpRms[b];
+#ifndef NO_DTX_MODE
+ if ((m_bitRateMode > 3) || (meanSpecFlat[ci] > (SCHAR_MAX >> 1)) || (rmsComp >= TA_EPS))
+#endif
if (rmsComp >= rmsRef9 && (rmsComp < (grpStepSizes[b] >> 1))) // zero-quantized
{
s -= ((grpOff[b + 1] - grpOff[b]) * reductionFactor * __min (2 * SA_EPS, rmsComp) + SA_EPS) >> 11; // / (2 * SA_EPS)
@@ -887,6 +892,9 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
const uint32_t rmsComp = getComplexRmsValue (grpRms[b], gr, b, numSwbCh, coreConfig.tnsData[ch]);
const uint32_t rmsRef9 = (!coreConfig.commonWindow ? rmsComp :
getComplexRmsValue (refRms[b], gr, b, numSwbCh, coreConfig.tnsData[1 - ch]) >> 9);
+#endif
+#ifndef NO_DTX_MODE
+ if ((m_bitRateMode > 3) || (meanSpecFlat[ci] > (SCHAR_MAX >> 1)) || (rmsComp >= TA_EPS))
#endif
if (rmsComp >= rmsRef9) // check only first SFB above max_sfb for simplification
{
@@ -901,7 +909,7 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
const unsigned lfAtten = (b <= 5 ? (eightShorts ? 1 : 4) + b * lfConst : 5 * lfConst - 1 + b + ((b + 5) >> 4));
const uint8_t sfbWidth = grpOff[b + 1] - grpOff[b];
const uint64_t rateFac = mSfmFac * s * __min (32, lfAtten * grpData.numWindowGroups); // rate control part 1
- const uint64_t sScaled = ((1u << 23) + __max (grpRmsMin, grpStepSizes[b]) * (scaleBr - (grpStereoData[b] > 0 ? 1 : 0)) * rateFac) >> 24;
+ const uint64_t sScaled = ((1u << 23) + __max (grpRmsMin, grpStepSizes[b]) * scaleBr * rateFac) >> 24;
// scale step-sizes according to VBR mode & derive scale factors from step-sizes
grpStepSizes[b] = uint32_t (__max (BA_EPS, __min (UINT_MAX, sScaled)));
@@ -919,6 +927,15 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
{
for (uint16_t gr = 0; gr < grpData.numWindowGroups; gr++)
{
+#ifndef NO_DTX_MODE
+ if (meanSpecFlat[ci] <= (SCHAR_MAX >> 1)) // noise-like low-energy "DTX" frame
+ {
+ for (s = (coreConfig.icsInfoPrev[ch].windowSequence == EIGHT_SHORT ? 24 : m_specGapFiller.getFirstGapFillSfb ()); s < maxSfbCh; s++)
+ {
+ if (grpData.sfbRmsValues[s + m_numSwbShort * gr] < ((3 * TA_EPS) >> 1)) grpData.scaleFactors[s + m_numSwbShort * gr] = 0;
+ }
+ }
+#endif
memset (&grpData.scaleFactors[maxSfbCh + m_numSwbShort * gr], 0, (numSwbFrame - maxSfbCh) * sizeof (uint8_t));
}
grpData.sfbsPerGroup = coreConfig.icsInfoCurr[ch].maxSfb = numSwbFrame;
@@ -1061,7 +1078,7 @@ unsigned ExhaleEncoder::quantizationCoding () // apply MDCT quantization and en
const uint8_t maxSfbLong = (samplingRate < 37566 ? 51 /*32 kHz*/ : brModeAndFsToMaxSfbLong (m_bitRateMode, samplingRate));
const uint8_t maxSfbShort = (samplingRate < 37566 ? 14 /*32 kHz*/ : brModeAndFsToMaxSfbShort(m_bitRateMode, samplingRate));
const uint16_t peakIndex = (shortWinCurr ? 0 : (m_specAnaCurr[ci] >> 5) & 2047);
- const unsigned sfmBasedSfbStart = (shortWinCurr ? maxSfbShort : maxSfbLong) - 5 + (m_bitRateMode >> 1) + (meanSpecFlat[ci] >> 5);
+ const unsigned sfmBasedSfbStart = (shortWinCurr ? maxSfbShort - 4 : maxSfbLong - 6) + (m_bitRateMode >> 1) + (meanSpecFlat[ci] >> 5);
const unsigned targetBitCount25 = ((60000 + 20000 * m_bitRateMode) * nSamplesInFrame) / (samplingRate * ((grpData.numWindowGroups + 1) >> 1));
unsigned b = grpData.sfbsPerGroup - 1;
diff --git a/src/lib/stereoProcessing.cpp b/src/lib/stereoProcessing.cpp
index 25d851e..3175b00 100644
--- a/src/lib/stereoProcessing.cpp
+++ b/src/lib/stereoProcessing.cpp
@@ -453,7 +453,7 @@ unsigned StereoProcessor::applyPredJointStereo (int32_t* const mdctSpectrum1, in
if (alphaRe == 0)
{
- if (realOnlyCalc) // update previous magnitude value
+ if (realOnlyCalc) // update previous res. MDCT value
{
sfbMdctR += sfbWidth - 1;
prevResi = (grpSData[sfbEv] > 0 ? *sfbMdctR : int32_t (((int64_t) sfbMdctD[sfbWidth - 1] +
|