TruncSilenceBase.cpp

Go to the documentation of this file.

/**********************************************************************
 
  Audacity: A Digital Audio Editor
 
  TruncSilenceBase.cpp
 
  Lynn Allan (from DM's Normalize)
  Philip Van Baren (more options and boundary fixes)
 
*******************************************************************//*******************************************************************/
#include "TruncSilenceBase.h"
#include "BasicUI.h"
#include "EffectOutputTracks.h"
#include "Prefs.h"
#include "Project.h"
#include "SyncLock.h"
#include "WaveTrack.h"
#include <algorithm>
#include <cmath>
#include <limits>
#include <list>
 
class Enums
{
public:
   static const size_t NumDbChoices;
   static const EnumValueSymbol DbChoices[];
};
 
const EnumValueSymbol Enums::DbChoices[] = {
   // Table of text values, only for reading what was stored in legacy config
   // files.
   // It was inappropriate to make this a discrete choice control.
   { wxT("-20 dB") }, { wxT("-25 dB") }, { wxT("-30 dB") }, { wxT("-35 dB") },
   { wxT("-40 dB") }, { wxT("-45 dB") }, { wxT("-50 dB") }, { wxT("-55 dB") },
   { wxT("-60 dB") }, { wxT("-65 dB") }, { wxT("-70 dB") }, { wxT("-75 dB") },
   { wxT("-80 dB") }
};
 
// Map from position in table above to numerical value.
static inline double enumToDB(int val)
{
   return -(5.0 * val + 20.0);
}
 
const size_t Enums::NumDbChoices = WXSIZEOF(Enums::DbChoices);
 
using Region = WaveTrack::Region;
 
// Declaration of RegionList
class RegionList : public std::list<Region>
{
};
 
const EnumValueSymbol TruncSilenceBase::kActionStrings[nActions] = {
   { XO("Truncate Detected Silence") }, { XO("Compress Excess Silence") }
};
 
static CommandParameters::ObsoleteMap kObsoleteActions[] = {
   // Compatible with 2.1.0 and before
   { wxT("0"), 0 }, // Remap to Truncate Detected Silence
   { wxT("1"), 1 }, // Remap to Compress Excess Silence
};
 
static const size_t nObsoleteActions = WXSIZEOF(kObsoleteActions);
 
const EffectParameterMethods& TruncSilenceBase::Parameters() const
{
   static CapturedParameters<
      TruncSilenceBase, Threshold, ActIndex, Minimum, Truncate, Compress,
      Independent>
      parameters;
   return parameters;
}
 
static const size_t DEF_BlendFrameCount = 100;
 
// Lower bound on the amount of silence to find at a time -- this avoids
// detecting silence repeatedly in low-frequency sounds.
static const double DEF_MinTruncMs = 0.001;
 
// Typical fraction of total time taken by detection (better to guess low)
const double detectFrac = 0.4;
 
const ComponentInterfaceSymbol TruncSilenceBase::Symbol { XO(
   "Truncate Silence") };
 
TruncSilenceBase::TruncSilenceBase()
{
   Parameters().Reset(*this);
 
   SetLinearEffectFlag(false);
 
   // This used to be changeable via the audacity.cfg/registry.  Doubtful that
   // was ever done.
   //
   // Original comment:
   //
   //   mBlendFrameCount only retrieved from prefs ... not using dialog
   //   Only way to change (for windows) is thru registry
   //   The values should be figured dynamically ... too many frames could be
   //   invalid
   mBlendFrameCount = DEF_BlendFrameCount;
}
 
TruncSilenceBase::~TruncSilenceBase()
{
}
 
// ComponentInterface implementation
 
ComponentInterfaceSymbol TruncSilenceBase::GetSymbol() const
{
   return Symbol;
}
 
TranslatableString TruncSilenceBase::GetDescription() const
{
   return XO(
      "Automatically reduces the length of passages where the volume is below a specified level");
}
 
ManualPageID TruncSilenceBase::ManualPage() const
{
   return L"Truncate_Silence";
}
 
// EffectDefinitionInterface implementation
 
EffectType TruncSilenceBase::GetType() const
{
   return EffectTypeProcess;
}
 
bool TruncSilenceBase::LoadSettings(
   const CommandParameters& parms, EffectSettings& settings) const
{
   Effect::LoadSettings(parms, settings);
 
   // A bit of special treatment for two parameters
 
   // This control migrated from a choice to a text box in version 2.3.0
   double myThreshold {};
   bool newParams = [&] {
      double temp;
      if (!parms.ReadAndVerify(
             Threshold.key, &temp, Threshold.def, Threshold.min, Threshold.max))
         return false;
      myThreshold = temp;
      return true;
   }();
 
   if (!newParams)
   {
      int temp;
      // Use legacy param:
      if (!parms.ReadAndVerify(
             L"Db", &temp, 0, Enums::DbChoices, Enums::NumDbChoices))
         return false;
      myThreshold = enumToDB(temp);
   }
 
   {
      int temp;
      if (!parms.ReadAndVerify(
             ActIndex.key, &temp, ActIndex.def, kActionStrings, nActions,
             kObsoleteActions, nObsoleteActions))
         return false;
 
      // TODO:  fix this when settings are really externalized
      const_cast<int&>(mActionIndex) = temp;
   }
   // TODO:  fix this when settings are really externalized
   const_cast<double&>(mThresholdDB) = myThreshold;
   return true;
}
 
// Effect implementation
 
double TruncSilenceBase::CalcPreviewInputLength(
   const EffectSettings&, double /* previewLength */) const
{
   double inputLength = mT1 - mT0;
   double minInputLength = inputLength;
 
   // Master list of silent regions
   RegionList silences;
 
   // Start with the whole selection silent
   silences.push_back(Region(mT0, mT1));
 
   int whichTrack = 0;
 
   for (auto wt : inputTracks()->Selected<const WaveTrack>())
   {
      RegionList trackSilences;
 
      auto index = wt->TimeToLongSamples(mT0);
      sampleCount silentFrame = 0; // length of the current silence
 
      Analyze(
         silences, trackSilences, *wt, &silentFrame, &index, whichTrack,
         &inputLength, &minInputLength);
 
      whichTrack += wt->NChannels();
   }
   return inputLength;
}
 
bool TruncSilenceBase::Process(EffectInstance&, EffectSettings&)
{
   const bool success = mbIndependent ? ProcessIndependently() : ProcessAll();
 
   return success;
}
 
bool TruncSilenceBase::ProcessIndependently()
{
   unsigned nGroups = 0;
 
   const bool syncLock = SyncLockState::Get(*FindProject()).IsSyncLocked();
 
   // Check if it's permissible
   {
      for (auto track : inputTracks()->Selected<const WaveTrack>())
      {
         if (syncLock)
         {
            auto otherTracks =
               SyncLock::Group(*track).Filter<const WaveTrack>() +
               &Track::IsSelected -
               [&](const Track* pTrack) { return pTrack == track; };
            if (otherTracks)
            {
               BasicUI::ShowMessageBox(XO(
                  "When truncating independently, there may only be one selected audio track in each Sync-Locked Track Group."));
               return false;
            }
         }
 
         ++nGroups;
      }
   }
 
   if (nGroups == 0)
      // nothing to do
      return true;
 
   // Now do the work
 
   // Copy tracks
   EffectOutputTracks outputs {
      *mTracks, GetType(), { { mT0, mT1 } }, true, true
   };
   double newT1 = 0.0;
 
   {
      unsigned iGroup = 0;
      for (auto track : outputs.Get().Selected<WaveTrack>())
      {
         RegionList silences;
         if (!FindSilences(
                silences, TrackList::SingletonRange(&as_const(*track))))
            return false;
         // Treat tracks in the sync lock group only
         Track *groupFirst, *groupLast;
         auto range = syncLock ? SyncLock::Group(*track) :
                                 TrackList::SingletonRange<Track>(track);
         double totalCutLen = 0.0;
         if (!DoRemoval(silences, range, iGroup, nGroups, totalCutLen))
            return false;
         newT1 = std::max(newT1, mT1 - totalCutLen);
 
         ++iGroup;
      }
   }
 
   mT1 = newT1;
 
   outputs.Commit();
 
   return true;
}
 
bool TruncSilenceBase::ProcessAll()
{
   // Copy tracks
   EffectOutputTracks outputs {
      *mTracks, GetType(), { { mT0, mT1 } }, true, true
   };
 
   // Master list of silent regions.
   // This list should always be kept in order.
   RegionList silences;
 
   if (FindSilences(silences, outputs.Get().Selected<const WaveTrack>()))
   {
      auto trackRange = outputs.Get().Any();
      double totalCutLen = 0.0;
      if (DoRemoval(silences, trackRange, 0, 1, totalCutLen))
      {
         mT1 -= totalCutLen;
         outputs.Commit();
         return true;
      }
   }
 
   return false;
}
 
bool TruncSilenceBase::FindSilences(
   RegionList& silences, const TrackIterRange<const WaveTrack>& range)
{
   // Start with the whole selection silent
   silences.push_back(Region(mT0, mT1));
 
   // Remove non-silent regions in each track
   int whichTrack = 0;
   for (auto wt : range)
   {
      // Smallest silent region to detect in frames
      auto minSilenceFrames = sampleCount(
         std::max(mInitialAllowedSilence, DEF_MinTruncMs) * wt->GetRate());
 
      //
      // Scan the track for silences
      //
      RegionList trackSilences;
 
      auto index = wt->TimeToLongSamples(mT0);
      sampleCount silentFrame = 0;
 
      // Detect silences
      bool cancelled = !(Analyze(
         silences, trackSilences, *wt, &silentFrame, &index, whichTrack));
 
      // Buffer has been freed, so we're OK to return if cancelled
      if (cancelled)
         return false;
 
      if (silentFrame >= minSilenceFrames)
      {
         // Track ended in silence -- record region
         trackSilences.push_back(Region(
            wt->LongSamplesToTime(index - silentFrame),
            wt->LongSamplesToTime(index)));
      }
 
      // Intersect with the overall silent region list
      Intersect(silences, trackSilences);
      whichTrack++;
   }
 
   return true;
}
 
bool TruncSilenceBase::DoRemoval(
   const RegionList& silences, const TrackIterRange<Track>& range,
   unsigned iGroup, unsigned nGroups, double& totalCutLen)
{
   //
   // Now remove the silent regions from all selected / sync-lock selected
   // tracks.
   //
 
   // Loop over detected regions in reverse (so cuts don't change time values
   // down the line)
   int whichReg = 0;
   RegionList::const_reverse_iterator rit;
   for (rit = silences.rbegin(); rit != silences.rend(); ++rit)
   {
      const Region& region = *rit;
      const Region* const r = &region;
 
      // Progress dialog and cancellation. Do additional cleanup before return.
      const double frac =
         detectFrac + (1 - detectFrac) *
                         (iGroup + whichReg / double(silences.size())) /
                         nGroups;
      if (TotalProgress(frac))
         return false;
 
      // Intersection may create regions smaller than allowed; ignore them.
      // Allow one nanosecond extra for consistent results with exact
      // milliseconds of allowed silence.
      if ((r->end - r->start) < (mInitialAllowedSilence - 0.000000001))
         continue;
 
      // Find NEW silence length as requested
      double inLength = r->end - r->start;
      double outLength;
 
      switch (mActionIndex)
      {
      case kTruncate:
         outLength = std::min(mTruncLongestAllowedSilence, inLength);
         break;
      case kCompress:
         outLength =
            mInitialAllowedSilence + (inLength - mInitialAllowedSilence) *
                                        mSilenceCompressPercent / 100.0;
         break;
      default: // Not currently used.
         outLength = std::min(
            mInitialAllowedSilence + (inLength - mInitialAllowedSilence) *
                                        mSilenceCompressPercent / 100.0,
            mTruncLongestAllowedSilence);
      }
 
      const double cutLen = std::max(0.0, inLength - outLength);
      // Don't waste time cutting nothing.
      if (cutLen == 0.0)
         continue;
 
      totalCutLen += cutLen;
 
      bool success = true;
      double cutStart = (r->start + r->end - cutLen) / 2;
      double cutEnd = cutStart + cutLen;
      (range + &SyncLock::IsSelectedOrSyncLockSelectedP -
       [&](const Track* pTrack) {
          return
             // Don't waste time past the end of a track
             pTrack->GetEndTime() < r->start;
       })
         .VisitWhile(
            success,
            [&](WaveTrack& wt) {
               // In WaveTracks, clear with a cross-fade
               auto blendFrames = mBlendFrameCount;
               // Round start/end times to frame boundaries
               cutStart = wt.SnapToSample(cutStart);
               cutEnd = wt.SnapToSample(cutEnd);
 
               // Make sure the cross-fade does not affect non-silent frames
               if (wt.LongSamplesToTime(blendFrames) > inLength)
               {
                  // Result is not more than blendFrames:
                  blendFrames = wt.TimeToLongSamples(inLength).as_size_t();
               }
 
               // Perform cross-fade in memory
               struct Buffers
               {
                  Buffers(size_t size)
                      : buf1 { size }
                      , buf2 { size }
                  {
                  }
                  Floats buf1, buf2;
               };
               Buffers buffers[2] { blendFrames, blendFrames };
               auto t1 = wt.TimeToLongSamples(cutStart) - blendFrames / 2;
               auto t2 = wt.TimeToLongSamples(cutEnd) - blendFrames / 2;
 
               size_t iChannel = 0;
               for (const auto pChannel : wt.Channels())
               {
                  auto& buffer = buffers[iChannel];
                  pChannel->GetFloats(buffer.buf1.get(), t1, blendFrames);
                  pChannel->GetFloats(buffer.buf2.get(), t2, blendFrames);
 
                  for (decltype(blendFrames) i = 0; i < blendFrames; ++i)
                  {
                     buffer.buf1[i] = ((blendFrames - i) * buffer.buf1[i] +
                                       i * buffer.buf2[i]) /
                                      (double)blendFrames;
                  }
                  ++iChannel;
               }
 
               wt.Clear(cutStart, cutEnd);
 
               iChannel = 0;
               for (const auto pChannel : wt.Channels())
               {
                  // Write cross-faded data
                  auto& buffer = buffers[iChannel];
                  success =
                     success &&
                     pChannel->SetFloats(
                        buffer.buf1.get(), t1, blendFrames,
                        // This effect mostly shifts samples to remove silences,
                        // and does only a little bit of floating point
                        // calculations to cross-fade the splices, over a 100
                        // sample interval by default. Don't dither.
                        narrowestSampleFormat);
                  ++iChannel;
               }
            },
            [&](Track& t) {
               // Non-wave tracks: just do a sync-lock adjust
               t.SyncLockAdjust(cutEnd, cutStart);
            });
      if (!success)
         return false;
      ++whichReg;
   }
 
   return true;
}
 
bool TruncSilenceBase::Analyze(
   RegionList& silenceList, RegionList& trackSilences, const WaveTrack& wt,
   sampleCount* silentFrame, sampleCount* index, int whichTrack,
   double* inputLength, double* minInputLength) const
{
   const auto rate = wt.GetRate();
 
   // Smallest silent region to detect in frames
   auto minSilenceFrames =
      sampleCount(std::max(mInitialAllowedSilence, DEF_MinTruncMs) * rate);
 
   double truncDbSilenceThreshold = DB_TO_LINEAR(mThresholdDB);
   auto blockLen = wt.GetMaxBlockSize();
   auto start = wt.TimeToLongSamples(mT0);
   auto end = wt.TimeToLongSamples(mT1);
   sampleCount outLength = 0;
 
   double previewLength;
   gPrefs->Read(wxT("/AudioIO/EffectsPreviewLen"), &previewLength, 6.0);
   // Minimum required length in samples.
   const sampleCount previewLen(previewLength * rate);
 
   // Keep position in overall silences list for optimization
   RegionList::iterator rit(silenceList.begin());
 
   // Allocate buffers
   Floats buffers[] { Floats { blockLen }, Floats { blockLen } };
 
   // Loop through current track
   while (*index < end)
   {
      if (
         inputLength &&
         ((outLength >= previewLen) ||
          (*index - start > wt.TimeToLongSamples(*minInputLength))))
      {
         *inputLength = std::min<double>(*inputLength, *minInputLength);
         if (outLength >= previewLen)
         {
            *minInputLength = *inputLength;
         }
         return true;
      }
 
      if (!inputLength)
      {
         // Show progress dialog, test for cancellation
         bool cancelled = TotalProgress(
            detectFrac *
            (whichTrack +
             (*index - start).as_double() / (end - start).as_double()) /
            (double)GetNumWaveTracks());
         if (cancelled)
            return false;
      }
 
      // Optimization: if not in a silent region skip ahead to the next one
 
      double curTime = wt.LongSamplesToTime(*index);
      for (; rit != silenceList.end(); ++rit)
      {
         // Find the first silent region ending after current time
         if (rit->end >= curTime)
         {
            break;
         }
      }
 
      if (rit == silenceList.end())
      {
         // No more regions -- no need to process the rest of the track
         if (inputLength)
         {
            // Add available samples up to previewLength.
            auto remainingTrackSamples =
               wt.TimeToLongSamples(wt.GetEndTime()) - *index;
            auto requiredTrackSamples = previewLen - outLength;
            outLength += (remainingTrackSamples > requiredTrackSamples) ?
                            requiredTrackSamples :
                            remainingTrackSamples;
         }
 
         break;
      }
      else if (rit->start > curTime)
      {
         // End current silent region, skip ahead
         if (*silentFrame >= minSilenceFrames)
         {
            trackSilences.push_back(Region(
               wt.LongSamplesToTime(*index - *silentFrame),
               wt.LongSamplesToTime(*index)));
         }
         *silentFrame = 0;
         auto newIndex = wt.TimeToLongSamples(rit->start);
         if (inputLength)
         {
            auto requiredTrackSamples = previewLen - outLength;
            // Add non-silent sample to outLength
            outLength += ((newIndex - *index) > requiredTrackSamples) ?
                            requiredTrackSamples :
                            newIndex - *index;
         }
 
         *index = newIndex;
      }
      // End of optimization
 
      // Limit size of current block if we've reached the end
      auto count = limitSampleBufferSize(blockLen, end - *index);
 
      // Fill buffers
      size_t iChannel = 0;
      for (const auto pChannel : wt.Channels())
         pChannel->GetFloats(buffers[iChannel++].get(), *index, count);
 
      // Look for silenceList in current block
      for (decltype(count) i = 0; i < count; ++i)
      {
         if (
            inputLength &&
            ((outLength >= previewLen) ||
             (outLength > wt.TimeToLongSamples(*minInputLength))))
         {
            *inputLength =
               wt.LongSamplesToTime(*index + i) - wt.LongSamplesToTime(start);
            break;
         }
 
         const bool silent =
            std::all_of(buffers, buffers + iChannel, [&](const Floats& buffer) {
               return fabs(buffer[i]) < truncDbSilenceThreshold;
            });
         if (silent)
            (*silentFrame)++;
         else
         {
            sampleCount allowed = 0;
            if (*silentFrame >= minSilenceFrames)
            {
               if (inputLength)
               {
                  switch (mActionIndex)
                  {
                  case kTruncate:
                     outLength +=
                        wt.TimeToLongSamples(mTruncLongestAllowedSilence);
                     break;
                  case kCompress:
                     allowed = wt.TimeToLongSamples(mInitialAllowedSilence);
                     outLength += sampleCount(
                        allowed.as_double() +
                        (*silentFrame - allowed).as_double() *
                           mSilenceCompressPercent / 100.0);
                     break;
                     // default: // Not currently used.
                  }
               }
 
               // Record the silent region
               trackSilences.push_back(Region(
                  wt.LongSamplesToTime(*index + i - *silentFrame),
                  wt.LongSamplesToTime(*index + i)));
            }
            else if (inputLength)
            { // included as part of non-silence
               outLength += *silentFrame;
            }
            *silentFrame = 0;
            if (inputLength)
            {
               ++outLength; // Add non-silent sample to outLength
            }
         }
      }
      // Next block
      *index += count;
   }
 
   if (inputLength)
   {
      *inputLength = std::min<double>(*inputLength, *minInputLength);
      if (outLength >= previewLen)
      {
         *minInputLength = *inputLength;
      }
   }
 
   return true;
}
 
// EffectTruncSilence implementation
 
// Finds the intersection of the ordered region lists, stores in dest
void TruncSilenceBase::Intersect(RegionList& dest, const RegionList& src)
{
   RegionList::iterator destIter;
   destIter = dest.begin();
   // Any time we reach the end of the dest list we're finished
   if (destIter == dest.end())
      return;
   RegionList::iterator curDest = destIter;
 
   // Operation: find non-silent regions in src, remove them from dest.
   double nsStart = curDest->start;
   double nsEnd;
   bool lastRun = false; // must run the loop one extra time
 
   RegionList::const_iterator srcIter = src.begin();
 
   // This logic, causing the loop to run once after end of src, must occur
   // each time srcIter is updated
   if (srcIter == src.end())
   {
      lastRun = true;
   }
 
   while (srcIter != src.end() || lastRun)
   {
      // Don't use curSrc unless lastRun is false!
      RegionList::const_iterator curSrc;
 
      if (lastRun)
      {
         // The last non-silent region extends as far as possible
         nsEnd = std::numeric_limits<double>::max();
      }
      else
      {
         curSrc = srcIter;
         nsEnd = curSrc->start;
      }
 
      if (nsEnd > nsStart)
      {
         // Increment through dest until we have a region that could be affected
         while (curDest->end <= nsStart)
         {
            ++destIter;
            if (destIter == dest.end())
            {
               return;
            }
            curDest = destIter;
         }
 
         // Check for splitting dest region in two
         if (nsStart > curDest->start && nsEnd < curDest->end)
         {
            // The second region
            Region r(nsEnd, curDest->end);
 
            // The first region
            curDest->end = nsStart;
 
            // Insert second region after first
            RegionList::iterator nextIt(destIter);
            ++nextIt;
 
            // This should just read: destIter = dest.insert(nextIt, r); but we
            // work around two two wxList::insert() bugs. First, in some
            // versions it returns the wrong value. Second, in some versions,
            // it crashes when you insert at list end.
            if (nextIt == dest.end())
               dest.push_back(r);
            else
               dest.insert(nextIt, r);
            ++destIter; // (now points at the newly-inserted region)
 
            curDest = destIter;
         }
 
         // Check for truncating the end of dest region
         if (
            nsStart > curDest->start && nsStart < curDest->end &&
            nsEnd >= curDest->end)
         {
            curDest->end = nsStart;
 
            ++destIter;
            if (destIter == dest.end())
            {
               return;
            }
            curDest = destIter;
         }
 
         // Check for all dest regions that need to be removed completely
         while (nsStart <= curDest->start && nsEnd >= curDest->end)
         {
            destIter = dest.erase(destIter);
            if (destIter == dest.end())
            {
               return;
            }
            curDest = destIter;
         }
 
         // Check for truncating the beginning of dest region
         if (
            nsStart <= curDest->start && nsEnd > curDest->start &&
            nsEnd < curDest->end)
         {
            curDest->start = nsEnd;
         }
      }
 
      if (lastRun)
      {
         // done
         lastRun = false;
      }
      else
      {
         // Next non-silent region starts at the end of this silent region
         nsStart = curSrc->end;
         ++srcIter;
         if (srcIter == src.end())
         {
            lastRun = true;
         }
      }
   }
}
 
/*
void TruncSilenceBase::BlendFrames(float* buffer, int blendFrameCount, int
leftIndex, int rightIndex)
{
   float* bufOutput = &buffer[leftIndex];
   float* bufBefore = &buffer[leftIndex];
   float* bufAfter  = &buffer[rightIndex];
   double beforeFactor = 1.0;
   double afterFactor  = 0.0;
   double adjFactor = 1.0 / (double)blendFrameCount;
   for (int j = 0; j < blendFrameCount; ++j)
   {
      bufOutput[j] = (float)((bufBefore[j] * beforeFactor) + (bufAfter[j] *
afterFactor)); beforeFactor -= adjFactor; afterFactor  += adjFactor;
   }
}
*/
 
bool TruncSilenceBase::NeedsDither() const
{
   return false;
}