fenix/exhale
1
0
Fork 0
mirror of https://gitlab.com/ecodis/exhale.git synced 2025-02-15 02:50:47 +01:00
Code Issues Packages Projects Releases Wiki Activity

tune MSE test code

Browse Source
This commit is contained in:
Christian R. Helmrich 2023-09-30 19:00:00 +02:00
parent 1b231994ec
commit 0b683be9c7
3 changed files with 59 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/app/exhaleApp.cpp
View File

@ -57,10 +57,16 @@
#define _GETCWD getcwd
#define _STRLEN strlen
#if 0 // change this to "#if 1" to avoid garbage text in some terminals
#define EXHALE_TEXT_INIT ""
#define EXHALE_TEXT_BLUE ""
#define EXHALE_TEXT_PINK ""
#else
#define EXHALE_TEXT_INIT "\x1b[0m"
#define EXHALE_TEXT_BLUE "\x1b[36m"
#define EXHALE_TEXT_PINK "\x1b[35m"
#endif
#endif
// constants, experimental macros
#if LE_ACCURATE_CALC
@ -866,7 +872,7 @@ int main (const int argc, char* argv[])
#ifdef NO_PREROLL_DATA
outAuData = (uint8_t*) malloc ((6144 >> 3) * numChannels); // max frame AU size
#else
outAuData = (uint8_t*) malloc ((9216 >> 3) * numChannels); // max frame AU size
outAuData = (uint8_t*) malloc ((9984 >> 3) * numChannels); // max frame AU size
#endif
if ((inPcmData == nullptr) || (outAuData == nullptr))
{

4
src/lib/bitStreamWriter.cpp
View File

@ -1,5 +1,5 @@
/* bitStreamWriter.cpp - source file for class with basic bit-stream writing capability
* written by C. R. Helmrich, last modified in 2021 - see License.htm for legal notices
* written by C. R. Helmrich, last modified in 2023 - see License.htm for legal notices
*
* The copyright in this software is being made available under the exhale Copyright License
* and comes with ABSOLUTELY NO WARRANTY. This software may be subject to other third-
@ -1135,7 +1135,7 @@ unsigned BitStreamWriter::createAudioFrame (CoreCoderData** const elementData,
}
else *rate = 0; // insufficient data
}
memcpy (accessUnit, &m_auBitStream.stream.front (), __min (ci * (ipf ? 1152 : 768), bitCount >> 3));
memcpy (accessUnit, &m_auBitStream.stream.front (), __min (ci * (ipf ? 1248 : 768), bitCount >> 3));
#endif
return (bitCount >> 3); // byte count
}

65
src/lib/exhaleEnc.cpp
View File

@ -288,6 +288,7 @@ static inline uint8_t brModeAndFsToMaxSfbShort(const unsigned bitRateMode, const
return (samplingRate > 51200 ? 11 : 13) - 2 + (bitRateMode >> 2);
}
#if !EE_MORE_MSE
static inline void findActualBandwidthShort (uint8_t* const maxSfbShort, const uint16_t* sfbOffsets,
const int32_t* mdctSignals, const int32_t* mdstSignals, const unsigned nSamplesInShort)
{
@ -322,6 +323,7 @@ static inline void findActualBandwidthShort (uint8_t* const maxSfbShort, const u
if (*maxSfbShort > maxSfb) *maxSfbShort = maxSfb;
}
#endif
static inline uint8_t stereoCorrGrouping (const SfbGroupData& grpData, const unsigned nSamplesInFrame, uint8_t* stereoCorrData)
{
@ -932,7 +934,9 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
#if !RESTRICT_TO_AAC
const uint8_t numSwbCh = (eightShorts ? m_numSwbShort : m_numSwbLong);
#endif
#if !EE_MORE_MSE
const uint16_t rateFac = m_bitAllocator.getRateCtrlFac (m_rateFactor, samplingRate, meanSpecFlat[ci], coreConfig.icsInfoPrev[ch].windowSequence == EIGHT_SHORT);
#endif
uint32_t* stepSizes = &sfbStepSizes[ci * m_numSwbShort * NUM_WINDOW_GROUPS];
memset (grpData.scaleFactors, 0, (MAX_NUM_SWB_SHORT * NUM_WINDOW_GROUPS) * sizeof (uint8_t));
@ -956,21 +960,30 @@ unsigned ExhaleEncoder::psychBitAllocation () // perceptual bit-allocation via s
{
s += unsigned (0.5 + sqrt ((double) abs (m_mdctSignals[1 - ci][b])));
}
s = (s + 1) >> 1;
s = (s + 1u) >> 1;
}
if (grpOff[maxSfbCh] > grpOff[0])
{
s = unsigned ((s * (eightShorts ? (24u + (grpData.windowGroupLength[gr] >> 2)) / grpData.windowGroupLength[gr] : 4u) + 4096u) >> 13);
s = unsigned ((s * (eightShorts ? (24u + (grpData.windowGroupLength[gr] >> 2)) / grpData.windowGroupLength[gr] : 3u) + 4096u) >> 13);
}
s = unsigned (__max (1u + (INT32_MAX >> ((eightShorts ? 1 : 2) + (2 + m_bitRateMode / 9) * m_bitRateMode)), s * s));
s = __max (1u + ((UINT32_MAX / (eightShorts ? 3u : 8u)) >> ((2 + m_bitRateMode / 9) * m_bitRateMode)), s * s);
#endif
for (unsigned b = 0; b < maxSfbCh; b++)
{
#if EE_MORE_MSE
const uint8_t sfbWidth = grpOff[b + 1] - grpOff[b];
const uint64_t sThresh = __max (1u + (INT32_MAX >> 29), (grpRms[b] * uint64_t (__max (16, b * b)) + 32u) >> 6);
const bool stereoCoded = (nrChannels == 2 && coreConfig.stereoMode > 0 && (coreConfig.stereoDataCurr[b] > 0 || !(coreConfig.stereoMode & 1)));
const uint32_t rmsbMax = (stereoCoded ? __max (grpRms[b], coreConfig.groupingData[1 - ch].sfbRmsValues[m_numSwbShort * gr + b]) : grpRms[b]);
const uint64_t sThresh = __max (1u + (UINT32_MAX >> 30), (rmsbMax * uint64_t (__max (16, b * b * grpData.numWindowGroups)) + 32u) >> 6);
const uint64_t predFac = (eightShorts || coreConfig.stereoMode < 3 || coreConfig.stereoDataCurr[b & 62] == 0 ? (eightShorts && !b ? 48u : 64u) :
uint64_t (0.5 + 64 - pow (__min (1.0, fabs (coreConfig.stereoDataCurr[b & 62] * 0.1 - 1.6)), 1.5) * 19.0)); // MS
grpStepSizes[b] = uint32_t (__min (sThresh, (s * predFac + 32u) >> 6));
if (stereoCoded && rmsbMax)
{
const uint32_t rmsCh = coreConfig.groupingData[1 - ch].sfbRmsValues[m_numSwbShort * gr + b];
grpStepSizes[b] = uint32_t (!eightShorts && s > sThresh ? sThresh : (eightShorts ? s >> __max (0, 2 - int (b)) : s));
grpStepSizes[b] = uint32_t (0.5 + grpStepSizes[b] * (1.0 - sqrt ((double) __min (grpRms[b], rmsCh) / rmsbMax) * 0.29289322));
}
#else
const unsigned lfConst = (samplingRate < 27713 && !eightShorts ? 1 : 2); // lfAtten: LF SNR boost, as in my M.Sc. thesis
const unsigned lfAtten = (b <= 5 ? (eightShorts ? 1 : 4) + b * lfConst : 5 * lfConst - 1 + b + ((b + 5) >> 4));
@ -1080,7 +1093,9 @@ unsigned ExhaleEncoder::quantizationCoding () // apply MDCT quantization and en
const unsigned nSamplesInFrame = toFrameLength (m_frameLength);
const unsigned samplingRate = toSamplingRate (m_frequencyIdx);
const unsigned nSamplesTempAna = (nSamplesInFrame * 25) >> 4; // pre-delay for look-ahead
#if !EE_MORE_MSE
const bool useMaxBandwidth = (samplingRate < 37566 || m_shiftValSBR > 0);
#endif
const unsigned* const coeffMagn = m_sfbQuantizer.getCoeffMagnPtr ();
uint8_t meanSpecFlat[USAC_MAX_NUM_CHANNELS];
uint8_t meanTempFlat[USAC_MAX_NUM_CHANNELS] = {208, 208, 208, 208, 208, 208, 208, 208};
@ -1183,14 +1198,14 @@ unsigned ExhaleEncoder::quantizationCoding () // apply MDCT quantization and en
if (grpData.sfbsPerGroup > 0) // rate control part 2 to reach constrained VBR (CVBR)
{
#if EE_MORE_MSE
const unsigned targetBitCount25 = INT32_MAX;
#else
const uint8_t maxSfbLong = (useMaxBandwidth ? 54 - (samplingRate >> 13) : brModeAndFsToMaxSfbLong (m_bitRateMode, samplingRate));
const uint8_t maxSfbShort = (useMaxBandwidth ? 19 - (samplingRate >> 13) : brModeAndFsToMaxSfbShort(m_bitRateMode, samplingRate));
const uint16_t peakIndex = (shortWinCurr ? 0 : (m_specAnaCurr[ci] >> 5) & 2047);
const unsigned sfmBasedSfbStart = (shortWinCurr ? maxSfbShort - 2 + (meanSpecFlat[ci] >> 6) : maxSfbLong - 6 + (meanSpecFlat[ci] >> 5)) +
(shortWinCurr ? -3 + (((1 << 5) + meanTempFlat[ci]) >> 6) : -7 + (((1 << 4) + meanTempFlat[ci]) >> 5));
#if EE_MORE_MSE
const unsigned targetBitCount25 = INT32_MAX;
#else
const unsigned targetBitCount25 = ((60000 + 20000 * ((m_bitRateMode + m_shiftValSBR) >> (m_frameCount <= 1 ? 2 : 0))) * nSamplesInFrame) /
(samplingRate * ((grpData.numWindowGroups + 1) >> 1));
#endif
@ -1475,7 +1490,9 @@ unsigned ExhaleEncoder::spectralProcessing () // complete ics_info(), calc TNS
}
icsCurr.maxSfb = __min (icsCurr.maxSfb, brModeAndFsToMaxSfbLong (m_bitRateMode, samplingRate));
}
#if !EE_MORE_MSE
while (grpSO[icsCurr.maxSfb] > __max (m_bandwidCurr[ci], m_bandwidPrev[ci]) + (icsCurr.maxSfb >> 1)) icsCurr.maxSfb--; // detect BW
#endif
}
else // icsCurr.windowSequence == EIGHT_SHORT
{
@ -1506,9 +1523,9 @@ unsigned ExhaleEncoder::spectralProcessing () // complete ics_info(), calc TNS
}
}
memcpy (grpData.windowGroupLength, windowGroupingTable[icsCurr.windowGrouping], NUM_WINDOW_GROUPS * sizeof (uint8_t));
#if !EE_MORE_MSE
findActualBandwidthShort (&icsCurr.maxSfb, grpSO, m_mdctSignals[ci], nChannels < 2 ? nullptr : m_mdstSignals[ci], nSamplesInShort);
#endif
errorValue |= eightShortGrouping (grpData, grpSO, m_mdctSignals[ci], nChannels < 2 ? nullptr : m_mdstSignals[ci]);
} // if EIGHT_SHORT
@ -1520,14 +1537,26 @@ unsigned ExhaleEncoder::spectralProcessing () // complete ics_info(), calc TNS
if (grpData.windowGroupLength[gr] == 1)
{
const uint8_t tonality = (m_specAnaCurr[ci] >> 16) & UCHAR_MAX;
#if EE_MORE_MSE
tnsData.filterOrder[n] = (m_bitRateMode >= EE_MORE_MSE ? 0 : m_linPredictor.calcOptTnsCoeffs (tnsData.coeffParCor[n], tnsData.coeff[n], &tnsData.coeffResLow[n],
tnsData.filterOrder[n], s, tonality >> (m_tempFlatPrev[ci] >> 5)));
#else
bool noTnsFilt = (m_bitRateMode >= EE_MORE_MSE || icsCurr.maxSfb <= 40);
if (!noTnsFilt && samplingRate >= 27713 && samplingRate < 55426 && icsCurr.maxSfb > 40)
{
errorValue |= m_specAnalyzer.getMeanAbsValues (m_mdctSignals[ci], m_mdstSignals[ci], nSamplesInFrame, ci, &grpSO[29], 12, grpData.sfbRmsValues);
if (errorValue == 0)
{
for (int b = 0; b < 12; b++)
{
errorValue += unsigned (0.5 + sqrt ((double) grpData.sfbRmsValues[b]));
}
noTnsFilt |= (errorValue < ((unsigned) m_bitRateMode << 7)); // avoid clicks
errorValue = 0;
}
}
if (noTnsFilt) tnsData.filterOrder[n] = 0; else
#endif
tnsData.filterOrder[n] = m_linPredictor.calcOptTnsCoeffs (tnsData.coeffParCor[n], tnsData.coeff[n], &tnsData.coeffResLow[n],
tnsData.filterOrder[n], s, tonality >> (m_tempFlatPrev[ci] >> 5));
#endif
tnsData.numFilters[n] = (tnsData.filterOrder[n] > 0 ? 1 : 0);
if ((ch == 0) && (icsCurr.windowSequence == EIGHT_SHORT) && (tnsData.numFilters[n] == 0) && (tnsData.firstTnsWindow == gr))
{
@ -1709,13 +1738,19 @@ unsigned ExhaleEncoder::temporalProcessing () // determine time-domain aspects o
const USAC_WSEQ wsPrev = icsPrev.windowSequence;
USAC_WSEQ& wsCurr = icsCurr.windowSequence;
// get temporal signal statistics, then determine overlap config. for the next frame
#if !EE_MORE_MSE
const unsigned plCurr = abs (m_tranLocCurr[ci]) & ((1024 << m_shiftValSBR) - 1);
#endif
const unsigned sfCurr = (m_tempAnaCurr[ci] >> 24) & UCHAR_MAX;
const unsigned tfCurr = (m_tempAnaCurr[ci] >> 16) & UCHAR_MAX;
#if !EE_MORE_MSE
const unsigned plNext = abs (m_tranLocNext[ci]) & ((1024 << m_shiftValSBR) - 1);
#endif
const unsigned sfNext = (m_tempAnaNext[ci] >> 24) & UCHAR_MAX;
const unsigned tfNext = (m_tempAnaNext[ci] >> 16) & UCHAR_MAX;
#if !EE_MORE_MSE
const unsigned tThresh = UCHAR_MAX * (__max (plCurr, plNext) < 614 /*0.6 * 1024*/ ? 16 : 15 - (m_bitRateMode >> 3));
#endif
tsCurr[ch] = (m_tempAnaCurr[ci] /*R*/) & UCHAR_MAX;
tsNext[ch] = (m_tempAnaNext[ci] >> 8) & UCHAR_MAX;