REFERENCE_TIME CTimeStretchFilter::DrainBuffers(IMediaSample* pSample, REFERENCE_TIME rtNewStart)
{
Log("TS - DrainBuffers - rtNewStart: %6.3f", rtNewStart / 10000000.0);
uint unprocessedSamplesBefore = numUnprocessedSamples();
uint zeros = flushEx() - 32; // Magic 32 to keep the SoundTouch's output in sync
uint unprocessedSamplesAfter = numUnprocessedSamples();
UINT32 outFramesAfter = numSamples();
UINT32 totalSamples = zeros + unprocessedSamplesBefore;
UINT32 totalProcessedSamples = totalSamples - unprocessedSamplesAfter;
Log("TS - DrainBuffers - unprocessedSamplesBefore: %u zeros: %u unprocessedSamplesAfter: %u outFramesAfter: %u duration %6.3f",
unprocessedSamplesBefore, zeros, unprocessedSamplesAfter, outFramesAfter, (double)unprocessedSamplesBefore * (double) UNITS / (double) m_pOutputFormat->Format.nSamplesPerSec);
REFERENCE_TIME rtAHwTime = 0;
REFERENCE_TIME rtRCTime = 0;
REFERENCE_TIME estimatedExtraSampleDuration = (((int)zeros - (int)unprocessedSamplesAfter) * UNITS) / m_pOutputFormat->Format.nSamplesPerSec;
double bias = m_pClock->GetBias();
double adjustment = m_pClock->Adjustment();
m_pClock->GetHWTime(&rtRCTime, &rtAHwTime);
double AVMult = m_pClock->SuggestedAudioMultiplier(rtAHwTime, rtRCTime, bias, adjustment);
setTempoInternal(AVMult, 1.0);
CreateOutput(totalProcessedSamples, outFramesAfter, bias, adjustment, AVMult, true);
// Empty SoundTouch's buffers
clear();
pSample->SetDiscontinuity(false);
return estimatedExtraSampleDuration;
}
HRESULT CTimeStretchFilter::CheckSample(IMediaSample* pSample)
{
if (!pSample)
return S_OK;
AM_MEDIA_TYPE *pmt = NULL;
bool bFormatChanged = false;
HRESULT hr = S_OK;
if (SUCCEEDED(pSample->GetMediaType(&pmt)) && pmt)
bFormatChanged = !FormatsEqual((WAVEFORMATEXTENSIBLE*)pmt->pbFormat, m_pInputFormat);
if (bFormatChanged)
{
uint unprocessedSamplesBefore = numUnprocessedSamples();
uint zeros = flushEx();
uint unprocessedSamplesAfter = numUnprocessedSamples();
UINT32 outFramesAfter = numSamples();
UINT32 totalSamples = zeros + unprocessedSamplesBefore;
UINT32 totalProcessedSamples = totalSamples - unprocessedSamplesAfter;
//double bias = (double)totalProcessedSamples / (double)outFramesAfter;
REFERENCE_TIME estimatedSampleDuration = totalProcessedSamples * UNITS / m_pOutputFormat->Format.nSamplesPerSec;
double bias = m_pClock->GetBias();
double adjustment = m_pClock->Adjustment();
double AVMult = m_pClock->SuggestedAudioMultiplier(estimatedSampleDuration, bias, adjustment);
setTempoInternal(AVMult, 1.0);
CreateOutput(totalProcessedSamples, outFramesAfter, bias, adjustment, AVMult, true);
// Empty SoundTouch's buffers
clear();
// Apply format change
ChannelOrder chOrder;
hr = NegotiateFormat((WAVEFORMATEXTENSIBLE*)pmt->pbFormat, 1, &chOrder);
pSample->SetDiscontinuity(false);
if (FAILED(hr))
{
DeleteMediaType(pmt);
Log("CTimeStretchFilter::CheckFormat failed to change format: 0x%08x", hr);
return hr;
}
else
{
m_chOrder = chOrder;
return S_FALSE; // format changed
}
}
return S_OK;
}
// Processing
DWORD CTimeStretchFilter::ThreadProc()
{
Log("CTimeStretchFilter::timestretch thread - starting up - thread ID: %d", m_ThreadId);
SetThreadName(0, "TimeStretchFilter");
AudioSinkCommand command;
CComPtr<IMediaSample> sample;
while (true)
{
m_csResources.Unlock();
HRESULT hr = GetNextSampleOrCommand(&command, &sample.p, INFINITE, &m_hSampleEvents, &m_dwSampleWaitObjects);
m_csResources.Lock();
if (hr == MPAR_S_THREAD_STOPPING)
{
Log("CTimeStretchFilter::timestretch thread - closing down - thread ID: %d", m_ThreadId);
SetEvent(m_hCurrentSampleReleased);
CloseThread();
m_csResources.Unlock();
return 0;
}
else
{
if (command == ASC_Flush)
{
Log("CTimeStretchFilter::timestretch thread - flushing");
m_rtInSampleTime = m_rtNextIncomingSampleTime = 0;
if (m_pNextOutSample)
m_pNextOutSample.Release();
flush();
sample.Release();
SetEvent(m_hCurrentSampleReleased);
}
else if (command == ASC_Pause || command == ASC_Resume)
continue;
else if (sample)
{
BYTE *pMediaBuffer = NULL;
long size = sample->GetActualDataLength();
if (sample->IsDiscontinuity() == S_OK)
{
sample->SetDiscontinuity(false);
m_bDiscontinuity = true;
}
if (CheckSample(sample) == S_FALSE)
{
DeleteMediaType(m_pMediaType);
sample->GetMediaType(&m_pMediaType);
}
CheckStreamContinuity(sample);
m_nSampleNum++;
hr = sample->GetPointer(&pMediaBuffer);
if ((hr == S_OK) && m_pMemAllocator)
{
uint unprocessedSamplesBefore = numUnprocessedSamples();
uint unprocessedSamplesAfter = 0;
UINT32 nFrames = size / m_pOutputFormat->Format.nBlockAlign;
REFERENCE_TIME estimatedSampleDuration = nFrames * UNITS / m_pOutputFormat->Format.nSamplesPerSec;
double bias = m_pClock->GetBias();
double adjustment = m_pClock->Adjustment();
double AVMult = m_pClock->SuggestedAudioMultiplier(estimatedSampleDuration, bias, adjustment);
setTempoInternal(AVMult, 1.0); // this should be the same as previous line, but in future we want to get rid of the 2nd parameter
// Process the sample
putSamplesInternal((const short*)pMediaBuffer, size / m_pOutputFormat->Format.nBlockAlign);
unprocessedSamplesAfter = numUnprocessedSamples();
UINT32 nInFrames = (size / m_pOutputFormat->Format.nBlockAlign) - unprocessedSamplesAfter + unprocessedSamplesBefore;
UINT32 nOutFrames = numSamples();
CreateOutput(nInFrames, nOutFrames, bias, adjustment, AVMult, false);
}
}
}
}
}
HRESULT CTimeStretchFilter::SetFormat(const WAVEFORMATEXTENSIBLE *pwfe)
{
std::vector<CSoundTouchEx*>* newStreams = NULL;
if (pwfe)
{
// First verify format is supported
if (pwfe->SubFormat != KSDATAFORMAT_SUBTYPE_PCM &&
pwfe->SubFormat != KSDATAFORMAT_SUBTYPE_IEEE_FLOAT)
return VFW_E_TYPE_NOT_ACCEPTED;
DWORD dwChannelMask = pwfe->dwChannelMask;
newStreams = new std::vector<CSoundTouchEx*>;
if (!newStreams)
return E_OUTOFMEMORY;
map<DWORD, int> inSpeakerOffset;
map<DWORD, int> outSpeakerOffset;
int currOffset = 0;
// Each bit position in dwChannelMask corresponds to a speaker position
// try every bit position from 0 to 31
for (DWORD dwSpeaker = 1; dwSpeaker != 0; dwSpeaker <<= 1)
{
if (dwChannelMask & dwSpeaker)
{
inSpeakerOffset[dwSpeaker] = currOffset;
currOffset += pwfe->Format.wBitsPerSample / 8;
}
}
ASSERT(inSpeakerOffset.size() == pwfe->Format.nChannels);
// PCM output, 1-to-1 mapping of input to output
outSpeakerOffset.insert(inSpeakerOffset.begin(), inSpeakerOffset.end());
// TODO: First create the base downmixing coefficients
// for syncing mono channels like LFE and Center
// Now start adding channels
// First try all speaker pairs
for (SpeakerPair *pPair = PairedSpeakers; pPair->dwLeft; pPair++)
{
if ((pPair->PairMask() & dwChannelMask) == pPair->PairMask())
{
CSoundTouchEx* pStream = new CSoundTouchEx();
pStream->setChannels(2);
pStream->SetInputChannels(inSpeakerOffset[pPair->dwLeft], inSpeakerOffset[pPair->dwRight]);
pStream->SetInputFormat(pwfe->Format.nBlockAlign, pwfe->Format.wBitsPerSample / 8);
pStream->SetOutputChannels(outSpeakerOffset[pPair->dwLeft], outSpeakerOffset[pPair->dwRight]);
pStream->SetOutputFormat(pwfe->Format.nBlockAlign, pwfe->Format.wBitsPerSample / 8);
pStream->SetupFormats();
newStreams->push_back(pStream);
dwChannelMask &= ~pPair->PairMask(); // mark channels as processed
}
}
// Then add all remaining channels as mono streams
// try every bit position from 0 to 31
for (DWORD dwSpeaker = 1; dwSpeaker != 0; dwSpeaker <<= 1)
{
if (dwChannelMask & dwSpeaker)
{
CSoundTouchEx *pStream = new CSoundTouchEx();
// TODO: make this a mixing stream, so that the channel can be synchronized
// to the mix of the main channels (normally Front Left/Right if available)
pStream->setChannels(1);
pStream->SetInputChannels(inSpeakerOffset[dwSpeaker]);
pStream->SetInputFormat(pwfe->Format.nBlockAlign, pwfe->Format.wBitsPerSample / 8);
pStream->SetOutputChannels(outSpeakerOffset[dwSpeaker]);
pStream->SetOutputFormat(pwfe->Format.nBlockAlign, pwfe->Format.wBitsPerSample / 8);
pStream->SetupFormats();
newStreams->push_back(pStream);
// The following is only necessary if we skip some channels
// currently we don't
//dwChannelMask &= ~dwSpeaker; // mark channel as processed
}
}
}
// delete old ones
std::vector<CSoundTouchEx*>* oldStreams = m_Streams;
m_Streams = newStreams;
if (oldStreams)
{
for (int i = 0; i < oldStreams->size(); i++)
{
SAFE_DELETE(oldStreams->at(i));
}
SAFE_DELETE(oldStreams);
}
if (m_Streams)
{
setTempoInternal(m_fNewTempo, m_fNewAdjustment);
setSampleRate(pwfe->Format.nSamplesPerSec);
}
return S_OK;
}
// Processing
DWORD CTimeStretchFilter::ThreadProc()
{
Log("CTimeStretchFilter::timestretch thread - starting up - thread ID: %d", m_ThreadId);
SetThreadName(0, "TimeStretchFilter");
AudioSinkCommand command;
CComPtr<IMediaSample> sample;
while (true)
{
m_csResources.Unlock();
HRESULT hr = GetNextSampleOrCommand(&command, &sample.p, INFINITE, &m_hSampleEvents, &m_dwSampleWaitObjects);
m_csResources.Lock();
if (hr == MPAR_S_THREAD_STOPPING)
{
Log("CTimeStretchFilter::timestretch thread - closing down - thread ID: %d", m_ThreadId);
SetEvent(m_hCurrentSampleReleased);
CloseThread();
m_csResources.Unlock();
return 0;
}
else
{
if (command == ASC_Flush)
{
Log("CTimeStretchFilter::timestretch thread - flushing");
m_rtInSampleTime = m_rtNextIncomingSampleTime = 0;
m_rtLastOuputStart = m_rtLastOuputEnd = -1;
if (m_pNextOutSample)
m_pNextOutSample.Release();
flush();
m_pClock->Flush();
sample.Release();
SetEvent(m_hCurrentSampleReleased);
}
else if (command == ASC_Pause || command == ASC_Resume)
continue;
else if (sample)
{
BYTE *pMediaBuffer = NULL;
long size = sample->GetActualDataLength();
if (sample->IsDiscontinuity() == S_OK)
{
sample->SetDiscontinuity(false);
m_bDiscontinuity = true;
}
REFERENCE_TIME rtDrained = 0;
if (CheckSample(sample, &rtDrained) == S_FALSE)
{
DeleteMediaType(m_pMediaType);
sample->GetMediaType(&m_pMediaType);
}
CheckStreamContinuity(sample, rtDrained);
m_nSampleNum++;
hr = sample->GetPointer(&pMediaBuffer);
if ((hr == S_OK) && m_pMemAllocator)
{
REFERENCE_TIME rtStart = 0;
REFERENCE_TIME rtAdjustedStart = 0;
REFERENCE_TIME rtEnd = 0;
REFERENCE_TIME rtAdjustedEnd = 0;
REFERENCE_TIME rtAHwTime = 0;
REFERENCE_TIME rtRCTime = 0;
m_pClock->GetHWTime(&rtRCTime, &rtAHwTime);
sample->GetTime(&rtStart, &rtEnd);
REFERENCE_TIME sampleDuration = rtEnd - rtStart;
uint unprocessedSamplesBefore = numUnprocessedSamples();
uint unprocessedSamplesAfter = 0;
UINT32 nFrames = size / m_pOutputFormat->Format.nBlockAlign;
double bias = m_pClock->GetBias();
double adjustment = m_pClock->Adjustment();
double AVMult = m_pClock->SuggestedAudioMultiplier(rtAHwTime, rtRCTime, bias, adjustment);
setTempoInternal(AVMult, 1.0);
if (m_rtLastOuputEnd == -1)
m_rtLastOuputEnd = rtStart / AVMult - 1;
m_rtLastOuputStart = m_rtLastOuputEnd + 1;
// Process the sample
putSamplesInternal((const short*)pMediaBuffer, size / m_pOutputFormat->Format.nBlockAlign);
unprocessedSamplesAfter = numUnprocessedSamples();
UINT32 nInFrames = (size / m_pOutputFormat->Format.nBlockAlign) - unprocessedSamplesAfter + unprocessedSamplesBefore;
UINT32 nOutFrames = numSamples();
// TODO: Soundtouch can provide less samples than asked (but never more) so a cummulative error is possible. This will not happen over the course of a long TV stint, but could be solved for correctness
// m_rtLastOuputEnd += (nOutFrames + unprocessedSamplesAfter - unprocessedSamplesBefore) * UNITS / m_pOutputFormat->Format.nSamplesPerSec;
//rtStart = m_rtInSampleTime;
rtEnd = rtStart + sampleDuration;
rtAdjustedStart = m_rtLastOuputEnd +1;
rtAdjustedEnd = rtAdjustedStart + sampleDuration / AVMult;
m_rtLastOuputEnd += sampleDuration / AVMult;
CreateOutput(nInFrames, nOutFrames, bias, adjustment, AVMult, false);
m_pClock->AddSample(rtStart, rtAdjustedStart, rtEnd, rtAdjustedEnd);
}
}
}
}
}
