This allows us to change the volume of sounds in filters ("volume",
"fade" properties) without quantizing the audio prematurely. Previously,
we had a hack to apply volume changes as a mixing-time step. This
can be done at any point in the filter chain now.
This also means we only need to worry about clipping at the very
last mixing step, instead of at every place we change sound.
Dithering for 16-bit audio and more intelligent gain control
is possible, as we have full-resolution, unclipped sound at the
mixing phase.
We support high resolution source sounds; both Vorbis and MAD
supply more than 16 bits.
Several filters, such as resampling, are easier to implement as
floats; these no longer need to convert back and forth.
Negatives:
- More memory use. The main case of this is in RageSoundReader_Preload,
which will be fixed: we can preload in 16-bit without losing most of the
above.
- Some extra overhead for accessing more memory.
351 lines
11 KiB
C++
351 lines
11 KiB
C++
#include "global.h"
|
|
#include "RageSoundReader_SpeedChange.h"
|
|
#include "RageSoundReader_Resample_Good.h"
|
|
#include "RageUtil.h"
|
|
#include "RageLog.h"
|
|
|
|
static const int WINDOW_SIZE_MS = 30;
|
|
|
|
RageSoundReader_SpeedChange::RageSoundReader_SpeedChange( RageSoundReader *pSource ):
|
|
RageSoundReader_Filter( pSource )
|
|
{
|
|
m_Channels.resize( pSource->GetNumChannels() );
|
|
m_fSpeedRatio = m_fTrailingSpeedRatio = 1.0f;
|
|
SetSpeedRatio( 1.0f );
|
|
Reset();
|
|
}
|
|
|
|
void RageSoundReader_SpeedChange::SetSpeedRatio( float fRatio )
|
|
{
|
|
m_fSpeedRatio = fRatio;
|
|
|
|
/* If we havn't read any data yet, put the new delta into effect immediately. */
|
|
if( m_iDataBufferAvailFrames == 0 )
|
|
m_fTrailingSpeedRatio = m_fSpeedRatio;
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::GetWindowSizeFrames() const
|
|
{
|
|
return (WINDOW_SIZE_MS * GetSampleRate()) / 1000;
|
|
}
|
|
|
|
void RageSoundReader_SpeedChange::Reset()
|
|
{
|
|
m_fTrailingSpeedRatio = m_fSpeedRatio;
|
|
m_iDataBufferAvailFrames = 0;
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
c.m_iCorrelatedPos = 0;
|
|
c.m_iLastCorrelatedPos = 0;
|
|
}
|
|
m_iUncorrelatedPos = 0;
|
|
m_iPos = 0;
|
|
m_fErrorFrames = 0;
|
|
}
|
|
|
|
static int FindClosestMatch( const float *pBuffer, int iBufferSize, const float *pCorrelateBuffer, int iCorrelateBufferSize, int iStride )
|
|
{
|
|
if( iBufferSize <= iCorrelateBufferSize )
|
|
return 0;
|
|
|
|
/* Optimization: if the buffers are the same, they always corellate at 0. */
|
|
if( pBuffer == pCorrelateBuffer )
|
|
return 0;
|
|
|
|
int iBufferDistanceToSearch = iBufferSize - iCorrelateBufferSize;
|
|
int iBestOffset = 0;
|
|
int fBestScore = 0;
|
|
for( int i = 0; i < iBufferDistanceToSearch; i += iStride )
|
|
{
|
|
float fScore = 0;
|
|
const float *pFrames = pBuffer + i;
|
|
for( int j = 0; j < iCorrelateBufferSize; j += iStride )
|
|
{
|
|
float fDiff = pFrames[j] - pCorrelateBuffer[j];
|
|
fScore += fabsf(fDiff);
|
|
}
|
|
|
|
if( i == 0 || fScore < fBestScore )
|
|
{
|
|
fBestScore = fScore;
|
|
iBestOffset = i;
|
|
}
|
|
}
|
|
return iBestOffset;
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::FillData( int iMaxFrames )
|
|
{
|
|
/* XXX: If the rate or source frame offset changes in the source, we should stop
|
|
* at that point, so the changes propagate upward. That's tricky, since we want
|
|
* to process a block at a time. We could just flush what we have, and maybe
|
|
* tweak m_iDeltaFrames for the next block based on how much we processed, so
|
|
* we average out the short block. */
|
|
while( iMaxFrames > 0 )
|
|
{
|
|
int iFramesToRead = iMaxFrames - m_iDataBufferAvailFrames;
|
|
int iBytesToRead = iFramesToRead * m_Channels.size() * sizeof(float);
|
|
if( iBytesToRead <= 0 )
|
|
return m_iDataBufferAvailFrames;
|
|
|
|
float *pTempBuffer = (float *) alloca( iBytesToRead );
|
|
int iGotFrames = m_pSource->Read( pTempBuffer, iFramesToRead );
|
|
if( iGotFrames < 0 )
|
|
{
|
|
if( iGotFrames == END_OF_FILE && m_iDataBufferAvailFrames )
|
|
return m_iDataBufferAvailFrames;
|
|
return iGotFrames;
|
|
}
|
|
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
|
|
if( (int) c.m_DataBuffer.size() < iMaxFrames )
|
|
c.m_DataBuffer.resize( iMaxFrames );
|
|
|
|
const float *pIn = pTempBuffer + i;
|
|
float *pOut = &c.m_DataBuffer[m_iDataBufferAvailFrames];
|
|
for( int j = 0; j < iGotFrames; ++j )
|
|
{
|
|
*pOut = *pIn;
|
|
pIn += m_Channels.size();
|
|
++pOut;
|
|
}
|
|
}
|
|
|
|
m_iDataBufferAvailFrames += iGotFrames;
|
|
}
|
|
return m_iDataBufferAvailFrames;
|
|
}
|
|
|
|
void RageSoundReader_SpeedChange::EraseData( int iFramesToDelete )
|
|
{
|
|
ASSERT( iFramesToDelete <= m_iDataBufferAvailFrames );
|
|
ASSERT( m_iDataBufferAvailFrames >= iFramesToDelete );
|
|
ASSERT( m_iUncorrelatedPos >= iFramesToDelete );
|
|
|
|
int iFramesToMove = m_iDataBufferAvailFrames - iFramesToDelete;
|
|
m_iDataBufferAvailFrames -= iFramesToDelete;
|
|
m_iUncorrelatedPos -= iFramesToDelete;
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
if( iFramesToMove )
|
|
memmove( &c.m_DataBuffer[0], &c.m_DataBuffer[iFramesToDelete], iFramesToMove * sizeof(float) );
|
|
ASSERT( c.m_iCorrelatedPos >= iFramesToDelete );
|
|
c.m_iCorrelatedPos -= iFramesToDelete;
|
|
}
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::Step()
|
|
{
|
|
if( m_iDataBufferAvailFrames == 0 )
|
|
return FillData( GetWindowSizeFrames() );
|
|
|
|
/* If m_iPos is non-zero, we just finished playing a previous block, so advance forward. */
|
|
if( m_iPos )
|
|
{
|
|
/* Advance m_iCorrelatedPos past the data that was just copied, to point to the
|
|
* sound that we would have played if we had continued copying at that point. */
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ASSERT( m_Channels[i].m_iCorrelatedPos + m_iPos <= m_iDataBufferAvailFrames );
|
|
m_Channels[i].m_iCorrelatedPos += m_iPos;
|
|
}
|
|
|
|
/* Advance m_iUncorrelatedPos to the position we'd prefer to continue playing from.
|
|
* This rounds the position, making the ratio imprecise. Record the amount of
|
|
* error introduced here. If we want to advance by 2.3 frames, we'll advance
|
|
* by 2.0 frames, and advance by 0.3 more the next time around. */
|
|
float fAdvanceFrames = GetWindowSizeFrames() * m_fTrailingSpeedRatio;
|
|
fAdvanceFrames += m_fErrorFrames;
|
|
int iTrailingDeltaFrames = lrintf( fAdvanceFrames );
|
|
m_fErrorFrames = fAdvanceFrames - iTrailingDeltaFrames;
|
|
m_iUncorrelatedPos += iTrailingDeltaFrames;
|
|
|
|
m_iPos = 0;
|
|
}
|
|
|
|
/* Update m_fTrailingSpeedRatio. Do this after advancing m_iUncorrelatedPos,
|
|
* so changes to m_iUncorrelatedPos line up with GetNextSourceFrame. */
|
|
m_fTrailingSpeedRatio = m_fSpeedRatio;
|
|
|
|
/* We don't need any data before the earlier of m_iUncorrelatedPos or m_iCorrelatedPos. */
|
|
int iToDelete = m_iUncorrelatedPos;
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
ASSERT( c.m_iCorrelatedPos <= m_iDataBufferAvailFrames );
|
|
iToDelete = min( iToDelete, c.m_iCorrelatedPos );
|
|
//iToDelete = min( iToDelete, m_iDataBufferAvailFrames );
|
|
}
|
|
EraseData( iToDelete );
|
|
|
|
/* Fill as much data as we might need to do the search and use the result. */
|
|
{
|
|
int iMaxPositionNeeded = m_iUncorrelatedPos + GetToleranceFrames() + GetWindowSizeFrames();
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
iMaxPositionNeeded = max( iMaxPositionNeeded, m_Channels[i].m_iCorrelatedPos + GetWindowSizeFrames() );
|
|
|
|
int iGot = FillData( iMaxPositionNeeded );
|
|
if( iGot < 0 )
|
|
return iGot;
|
|
|
|
if( iMaxPositionNeeded > m_iDataBufferAvailFrames )
|
|
{
|
|
/* We're at EOF. Flush the remaining data, if any. */
|
|
m_iUncorrelatedPos = m_Channels[0].m_iCorrelatedPos;
|
|
return m_iDataBufferAvailFrames;
|
|
}
|
|
}
|
|
|
|
/* Starting at our preferred position (m_iUncorrelatedPos), within GetToleranceFrames(),
|
|
* search for data that looks like the sound we would have continued playing if we
|
|
* kept going from the old position (m_iCorrelatedPos). */
|
|
int iCorrelatedToMatch = GetWindowSizeFrames()/4;
|
|
int iUncorrelatedToMatch = GetToleranceFrames() + iCorrelatedToMatch; // maximum distance to search
|
|
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
ASSERT( c.m_iCorrelatedPos >= 0 );
|
|
ASSERT( c.m_iCorrelatedPos < m_iDataBufferAvailFrames );
|
|
|
|
int iBest = FindClosestMatch( &c.m_DataBuffer[m_iUncorrelatedPos], iUncorrelatedToMatch, &c.m_DataBuffer[c.m_iCorrelatedPos], iCorrelatedToMatch, m_Channels.size() );
|
|
c.m_iLastCorrelatedPos = c.m_iCorrelatedPos;
|
|
c.m_iCorrelatedPos = iBest + m_iUncorrelatedPos;
|
|
ASSERT( m_Channels[i].m_iCorrelatedPos + GetWindowSizeFrames() <= m_iDataBufferAvailFrames );
|
|
}
|
|
return m_iDataBufferAvailFrames;
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::GetCursorAvail() const
|
|
{
|
|
int iCursorAvail = GetWindowSizeFrames() - m_iPos;
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
int iCursorAvailForChannel = (m_iDataBufferAvailFrames-m_Channels[i].m_iCorrelatedPos) - m_iPos;
|
|
iCursorAvail = min( iCursorAvail, iCursorAvailForChannel );
|
|
}
|
|
|
|
return iCursorAvail;
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::Read( float *pBuf, int iFrames )
|
|
{
|
|
while( 1 )
|
|
{
|
|
if( m_iDataBufferAvailFrames == 0 && m_fTrailingSpeedRatio == m_fSpeedRatio && m_fSpeedRatio == 1.0f )
|
|
{
|
|
/* Fast path: the buffer is empty, and we're not scaling the audio. Read directly
|
|
* into the output buffer, to eliminate memory and copying overhead. */
|
|
return m_pSource->Read( pBuf, iFrames );
|
|
}
|
|
|
|
int iCursorAvail = GetCursorAvail();
|
|
|
|
if( iCursorAvail == 0 )
|
|
{
|
|
int iRet = Step();
|
|
if( iRet < 0 )
|
|
return iRet;
|
|
if( !GetCursorAvail() )
|
|
return END_OF_FILE;
|
|
continue;
|
|
}
|
|
|
|
/* copy GetWindowSizeFrames() from iCorrelatedPos */
|
|
int iFramesLen = iFrames;
|
|
int iFramesAvail = min( iCursorAvail, iFramesLen );
|
|
|
|
iFrames -= iFramesAvail;
|
|
int iFramesRead = iFramesAvail;
|
|
|
|
int iWindowSizeFrames = GetWindowSizeFrames();
|
|
while( iFramesAvail-- )
|
|
{
|
|
for( size_t i = 0; i < m_Channels.size(); ++i )
|
|
{
|
|
ChannelInfo &c = m_Channels[i];
|
|
float i1 = c.m_DataBuffer[c.m_iCorrelatedPos+m_iPos];
|
|
float i2 = c.m_DataBuffer[c.m_iLastCorrelatedPos+m_iPos];
|
|
*pBuf++ = SCALE( m_iPos, 0, iWindowSizeFrames, i2, i1 );
|
|
}
|
|
|
|
++m_iPos;
|
|
}
|
|
|
|
return iFramesRead;
|
|
}
|
|
}
|
|
|
|
/* We prefer to be able to seek precisely, so seeking to a position produces data
|
|
* equal to what you'd get if you read data up to that point. This filter can't do
|
|
* that, because the exact selection of slices is dependent on the previous selection. */
|
|
int RageSoundReader_SpeedChange::SetPosition( int iFrame )
|
|
{
|
|
Reset();
|
|
return RageSoundReader_Filter::SetPosition( iFrame );
|
|
}
|
|
|
|
bool RageSoundReader_SpeedChange::SetProperty( const RString &sProperty, float fValue )
|
|
{
|
|
if( sProperty == "Speed" )
|
|
{
|
|
SetSpeedRatio( fValue );
|
|
return true;
|
|
}
|
|
|
|
return RageSoundReader_Filter::SetProperty( sProperty, fValue );
|
|
}
|
|
|
|
int RageSoundReader_SpeedChange::GetNextSourceFrame() const
|
|
{
|
|
float fRatio = m_fTrailingSpeedRatio;
|
|
|
|
int iSourceFrame = RageSoundReader_Filter::GetNextSourceFrame();
|
|
int iPos = lrintf(m_iPos * fRatio);
|
|
|
|
iSourceFrame -= m_iDataBufferAvailFrames;
|
|
iSourceFrame += m_iUncorrelatedPos + iPos;
|
|
return iSourceFrame;
|
|
}
|
|
|
|
float RageSoundReader_SpeedChange::GetStreamToSourceRatio() const
|
|
{
|
|
/* If the ratio has changed and GetCursorAvail() is 0, the new ratio will
|
|
* become the real ratio on the next read. Otherwise, we'll continue using
|
|
* our old ratio for a bit longer. */
|
|
if( GetCursorAvail() == 0 )
|
|
return m_fSpeedRatio * RageSoundReader_Filter::GetStreamToSourceRatio();
|
|
else
|
|
return m_fTrailingSpeedRatio * RageSoundReader_Filter::GetStreamToSourceRatio();
|
|
}
|
|
|
|
/*
|
|
* Copyright (c) 2006 Glenn Maynard
|
|
* All rights reserved.
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
* copy of this software and associated documentation files (the
|
|
* "Software"), to deal in the Software without restriction, including
|
|
* without limitation the rights to use, copy, modify, merge, publish,
|
|
* distribute, and/or sell copies of the Software, and to permit persons to
|
|
* whom the Software is furnished to do so, provided that the above
|
|
* copyright notice(s) and this permission notice appear in all copies of
|
|
* the Software and that both the above copyright notice(s) and this
|
|
* permission notice appear in supporting documentation.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
|
|
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT OF
|
|
* THIRD PARTY RIGHTS. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR HOLDERS
|
|
* INCLUDED IN THIS NOTICE BE LIABLE FOR ANY CLAIM, OR ANY SPECIAL INDIRECT
|
|
* OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
|
|
* OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
|
|
* OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
|
|
* PERFORMANCE OF THIS SOFTWARE.
|
|
*/
|