/// <summary>
/// Initialize Kinect audio stream object.
/// </summary>
/// <returns>
/// <para>S_OK on success, otherwise failure code.</para>
/// </returns>
HRESULT KinectReader::InitializeAudioStream()
{
INuiAudioBeam* pNuiAudioSource = NULL;
IMediaObject* pDMO = NULL;
IPropertyStore* pPropertyStore = NULL;
IStream* pStream = NULL;
// Get the audio source
HRESULT hr = m_pNuiSensor->NuiGetAudioSource(&pNuiAudioSource);
if (SUCCEEDED(hr))
{
hr = pNuiAudioSource->QueryInterface(IID_IMediaObject, (void**)&pDMO);
if (SUCCEEDED(hr))
{
hr = pNuiAudioSource->QueryInterface(IID_IPropertyStore, (void**)&pPropertyStore);
// Set AEC-MicArray DMO system mode. This must be set for the DMO to work properly.
// Possible values are:
// SINGLE_CHANNEL_AEC = 0
// OPTIBEAM_ARRAY_ONLY = 2
// OPTIBEAM_ARRAY_AND_AEC = 4
// SINGLE_CHANNEL_NSAGC = 5
PROPVARIANT pvSysMode;
PropVariantInit(&pvSysMode);
pvSysMode.vt = VT_I4;
pvSysMode.lVal = (LONG)(2); // Use OPTIBEAM_ARRAY_ONLY setting. Set OPTIBEAM_ARRAY_AND_AEC instead if you expect to have sound playing from speakers.
pPropertyStore->SetValue(MFPKEY_WMAAECMA_SYSTEM_MODE, pvSysMode);
PropVariantClear(&pvSysMode);
// Set DMO output format
WAVEFORMATEX wfxOut = {AudioFormat, AudioChannels, AudioSamplesPerSecond, AudioAverageBytesPerSecond, AudioBlockAlign, AudioBitsPerSample, 0};
DMO_MEDIA_TYPE mt = {0};
MoInitMediaType(&mt, sizeof(WAVEFORMATEX));
mt.majortype = MEDIATYPE_Audio;
mt.subtype = MEDIASUBTYPE_PCM;
mt.lSampleSize = 0;
mt.bFixedSizeSamples = TRUE;
mt.bTemporalCompression = FALSE;
mt.formattype = FORMAT_WaveFormatEx;
memcpy(mt.pbFormat, &wfxOut, sizeof(WAVEFORMATEX));
hr = pDMO->SetOutputType(0, &mt, 0);
if (SUCCEEDED(hr))
{
m_pKinectAudioStream = new KinectAudioStream(pDMO);
hr = m_pKinectAudioStream->QueryInterface(IID_IStream, (void**)&pStream);
if (SUCCEEDED(hr))
{
hr = CoCreateInstance(CLSID_SpStream, NULL, CLSCTX_INPROC_SERVER, __uuidof(ISpStream), (void**)&m_pSpeechStream);
if (SUCCEEDED(hr))
{
hr = m_pSpeechStream->SetBaseStream(pStream, SPDFID_WaveFormatEx, &wfxOut);
}
}
}
MoFreeMediaType(&mt);
}
}
SafeRelease(pStream);
SafeRelease(pPropertyStore);
SafeRelease(pDMO);
SafeRelease(pNuiAudioSource);
return hr;
}
HRESULT
KinectImpl::initAudio(){
HRESULT hr = S_OK;
// Set high priority to avoid getting preempted while capturing sound
#if 1
DWORD m_TaskIndex=0;
h_TaskIndex = AvSetMmThreadCharacteristicsW(L"Audio", &m_TaskIndex);
if (h_TaskIndex == NULL) {
hr = E_FAIL;
puts("Failed to set thread priority\n");
return hr;
}
#endif
// DMO initialization
INuiAudioBeam* pAudio;
hr = m_pNuiSensor->NuiGetAudioSource(&pAudio);
if (FAILED(hr)) {
puts("Failed to NuiGetAudioSource\n");
return hr;
}
hr = pAudio->QueryInterface(IID_IMediaObject, (void**)&pMediaObject);
if (FAILED(hr)) {
puts("Fail to call pAudio->QueryInterface(pMediaObject)\n");
return hr;
}
hr = pAudio->QueryInterface(IID_IPropertyStore, (void**)&pPropertyStore);
if (FAILED(hr)) {
puts("Fail to call pAudio->QueryInterface(pPropertyStore)\n");
return hr;
}
pAudio->Release();
// Set AEC-MicArray DMO system mode.
// This must be set for the DMO to work properly
PROPVARIANT pvSysMode;
PropVariantInit(&pvSysMode);
pvSysMode.vt = VT_I4;
pvSysMode.lVal = (LONG)(4);
hr = pPropertyStore->SetValue(MFPKEY_WMAAECMA_SYSTEM_MODE, pvSysMode);
if (FAILED(hr)) {
puts("Fail to call pPS->SetValue\n");
return hr;
}
PropVariantClear(&pvSysMode);
// NOTE: Need to wait 4 seconds for device to be ready right after initialization
DWORD dwWait = 4;
while (dwWait > 0){
_tprintf(_T("Device will be ready for recording in %d second(s).\r"), dwWait--);
Sleep(1000);
}
_tprintf(_T("Device is ready. \n"));
/*** set output buffer ***/
memset(&OutputBufferStruct,0, sizeof(OutputBufferStruct));
OutputBufferStruct.pBuffer = &outputBuffer;
// Set DMO output format
WAVEFORMATEX wfxOut = {WAVE_FORMAT_PCM, 1, 16000, 32000, 2, 16, 0};
DMO_MEDIA_TYPE mt = {0};
hr = MoInitMediaType(&mt, sizeof(WAVEFORMATEX));
if (FAILED(hr)) {
puts("MoInitMediaType failed\n");
return hr;
}
mt.majortype = MEDIATYPE_Audio;
mt.subtype = MEDIASUBTYPE_PCM;
mt.lSampleSize = 0;
mt.bFixedSizeSamples = TRUE;
mt.bTemporalCompression = FALSE;
mt.formattype = FORMAT_WaveFormatEx;
memcpy(mt.pbFormat, &wfxOut, sizeof(WAVEFORMATEX));
hr = pMediaObject->SetOutputType(0, &mt, 0);
if (FAILED(hr)) {
puts("SetOutputType failed\n");
return hr;
}
MoFreeMediaType(&mt);
hr = pMediaObject->AllocateStreamingResources();
if (FAILED(hr)) {
puts("AllocateStreamingResources failed\n");
return hr;
}
int iFrameSize;
PROPVARIANT pvFrameSize;
PropVariantInit(&pvFrameSize);
hr = pPropertyStore->GetValue(MFPKEY_WMAAECMA_FEATR_FRAME_SIZE, &pvFrameSize);
if (FAILED(hr)) {
puts("Fail to call pPS->GetValue\n");
return hr;
}
iFrameSize = pvFrameSize.lVal;
PropVariantClear(&pvFrameSize);
// allocate output buffer
DWORD cOutputBufLen = wfxOut.nSamplesPerSec * wfxOut.nBlockAlign;
pbOutputBuffer = new BYTE[cOutputBufLen];
if (pbOutputBuffer == NULL) {
puts("Fail to allocate output buffer.\n");
return hr;
}
//
hr = pMediaObject->QueryInterface(IID_INuiAudioBeam, (void**)&pAudioBeam);
if (FAILED(hr)) {
puts("QueryInterface for IID_INuiAudioBeam failed\n");
return hr;
}
outputBuffer.Init((byte*)pbOutputBuffer, cOutputBufLen, 0);
return S_OK;
}
int _tmain( int argc, _TCHAR* argv[] )
{
cv::setUseOptimized( true );
// Kinectのインスタンス生成、初期化
INuiSensor* pSensor;
HRESULT hResult = S_OK;
hResult = NuiCreateSensorByIndex( 0, &pSensor );
if( FAILED( hResult ) ){
std::cerr << "Error : NuiCreateSensorByIndex" << std::endl;
return -1;
}
hResult = pSensor->NuiInitialize( NUI_INITIALIZE_FLAG_USES_AUDIO );
if( FAILED( hResult ) ){
std::cerr << "Error : NuiInitialize" << std::endl;
return -1;
}
// Audioストリームの初期化(InitializeAudioStream)
std::cout << "InitializeAudioStream" << std::endl;
INuiAudioBeam* pNuiAudioSource;
hResult = pSensor->NuiGetAudioSource( &pNuiAudioSource );
if( FAILED( hResult ) ){
std::cerr << "Error : NuiGetAudioSource" << std::endl;
return -1;
}
IMediaObject* pMediaObject = nullptr;
IPropertyStore* pPropertyStore = nullptr;
pNuiAudioSource->QueryInterface( IID_IMediaObject, reinterpret_cast<void**>( &pMediaObject ) );
pNuiAudioSource->QueryInterface( IID_IPropertyStore, reinterpret_cast<void**>( &pPropertyStore ) );
PROPVARIANT propvariant;
PropVariantInit( &propvariant );
propvariant.vt = VT_I4;
propvariant.lVal = static_cast<LONG>( 4 );
pPropertyStore->SetValue( MFPKEY_WMAAECMA_SYSTEM_MODE, propvariant );
PropVariantClear( &propvariant );
WAVEFORMATEX waveFormat = { AudioFormat, AudioChannels, AudioSamplesPerSecond, AudioAverageBytesPerSecond, AudioBlockAlign, AudioBitsPerSample, 0 };
DMO_MEDIA_TYPE mediaType = { 0 };
MoInitMediaType( &mediaType, sizeof( WAVEFORMATEX ) );
mediaType.majortype = MEDIATYPE_Audio;
mediaType.subtype = MEDIASUBTYPE_PCM;
mediaType.lSampleSize = 0;
mediaType.bFixedSizeSamples = true;
mediaType.bTemporalCompression = false;
mediaType.formattype = FORMAT_WaveFormatEx;
memcpy( mediaType.pbFormat, &waveFormat, sizeof( WAVEFORMATEX ) );
pMediaObject->SetOutputType( 0, &mediaType, 0 );
KinectAudioStream* audioStream = new KinectAudioStream( pMediaObject );
IStream* pStream = nullptr;
audioStream->QueryInterface( IID_IStream, reinterpret_cast<void**>( &pStream ) );
CoInitialize( nullptr );
ISpStream* pSpeechStream = nullptr;
CoCreateInstance( CLSID_SpStream, NULL, CLSCTX_INPROC_SERVER, __uuidof(ISpStream), reinterpret_cast<void**>( &pSpeechStream ) );
pSpeechStream->SetBaseStream( pStream, SPDFID_WaveFormatEx, &waveFormat );
MoFreeMediaType( &mediaType );
pStream->Release();
pPropertyStore->Release();
pMediaObject->Release();
pNuiAudioSource->Release();
// 音声認識器を作成(CreateSpeechRecognizer)
std::cout << "CreateSpeechRecognizer" << std::endl;
ISpRecognizer* pSpeechRecognizer;
CoCreateInstance( CLSID_SpInprocRecognizer, nullptr, CLSCTX_INPROC_SERVER, __uuidof(ISpRecognizer), reinterpret_cast<void**>( &pSpeechRecognizer ) );
pSpeechRecognizer->SetInput( pSpeechStream, false );
/*
// If can use ATL, easier to using SpFindBestToken(sphelper.h). When using Professional or more.
ISpObjectToken* pEngineToken = nullptr;
SpFindBestToken( SPCAT_RECOGNIZERS, L"Language=411;Kinect=True", NULL, &pEngineToken ); // Japanese "Language=411;Kinect=True" English "Language=409;Kinect=True"
*/
///*
// If can't use ATL, alternative to using SpFIndBestToken(sphelper.h). When using Express.
const wchar_t* pVendorPreferred = L"VendorPreferred";
const unsigned long lengthVendorPreferred = static_cast<unsigned long>( wcslen( pVendorPreferred ) );
unsigned long length;
ULongAdd( lengthVendorPreferred, 1, &length );
wchar_t* pAttribsVendorPreferred = new wchar_t[ length ];
StringCchCopyW( pAttribsVendorPreferred, length, pVendorPreferred );
ISpObjectTokenCategory* pTokenCategory = nullptr;
CoCreateInstance( CLSID_SpObjectTokenCategory, nullptr, CLSCTX_ALL, __uuidof(ISpObjectTokenCategory), reinterpret_cast<void**>( &pTokenCategory ) );
pTokenCategory->SetId( SPCAT_RECOGNIZERS, false );
IEnumSpObjectTokens* pEnumTokens = nullptr;
CoCreateInstance( CLSID_SpMMAudioEnum, nullptr, CLSCTX_ALL, __uuidof(IEnumSpObjectTokens), reinterpret_cast<void**>( &pEnumTokens ) );
pTokenCategory->EnumTokens( L"Language=411;Kinect=True", pAttribsVendorPreferred, &pEnumTokens ); // Japanese "Language=411;Kinect=True" English "Language=409;Kinect=True"
delete[] pAttribsVendorPreferred;
ISpObjectToken* pEngineToken = nullptr;
pEnumTokens->Next( 1, &pEngineToken, nullptr );
//*/
pSpeechRecognizer->SetRecognizer( pEngineToken );
ISpRecoContext* pSpeechContext;
pSpeechRecognizer->CreateRecoContext( &pSpeechContext );
pEngineToken->Release();
///*
pTokenCategory->Release();
pEnumTokens->Release();
//*/
// 音声認識辞書の作成(LoadSpeechGrammar)
std::cout << "LoadSpeechGrammar" << std::endl;
ISpRecoGrammar* pSpeechGrammar;
pSpeechContext->CreateGrammar( 1, &pSpeechGrammar );
pSpeechGrammar->LoadCmdFromFile( L"SpeechRecognition_Ja.grxml", /*SPLO_STATIC*/SPLO_DYNAMIC ); // http://www.w3.org/TR/speech-grammar/ (UTF-8/CRLF)
audioStream->StartCapture();
pSpeechGrammar->SetRuleState( nullptr, nullptr, SPRS_ACTIVE );
pSpeechRecognizer->SetRecoState( SPRST_ACTIVE_ALWAYS );
pSpeechContext->SetInterest( SPFEI( SPEI_RECOGNITION ), SPFEI( SPEI_RECOGNITION ) );
pSpeechContext->Resume( 0 );
HANDLE hSpeechEvent = INVALID_HANDLE_VALUE;
hSpeechEvent = pSpeechContext->GetNotifyEventHandle();
HANDLE hEvents[1] = { hSpeechEvent };
int width = 640;
int height = 480;
cv::Mat audioMat = cv::Mat::zeros( height, width, CV_8UC3 );
cv::namedWindow( "Audio" );
bool exit = false;
std::cout << std::endl << "Speech Recognition Start..." << std::endl << std::endl;
while( 1 ){
// イベントの更新待ち
ResetEvent( hSpeechEvent );
unsigned long waitObject = MsgWaitForMultipleObjectsEx( ARRAYSIZE( hEvents ), hEvents, INFINITE, QS_ALLINPUT, MWMO_INPUTAVAILABLE );
if( waitObject == WAIT_OBJECT_0 ){
// イベントの取得
const float confidenceThreshold = 0.3f;
SPEVENT eventStatus;
unsigned long eventFetch = 0;
pSpeechContext->GetEvents( 1, &eventStatus, &eventFetch );
while( eventFetch > 0 ){
switch( eventStatus.eEventId ){
// 音声認識イベント(SPEI_HYPOTHESIS:推定またはSPEI_RECOGNITION:認識)
case SPEI_HYPOTHESIS:
case SPEI_RECOGNITION:
if( eventStatus.elParamType == SPET_LPARAM_IS_OBJECT ){
// フレーズの取得
ISpRecoResult* pRecoResult = reinterpret_cast<ISpRecoResult*>( eventStatus.lParam );
SPPHRASE* pPhrase = nullptr;
hResult = pRecoResult->GetPhrase( &pPhrase );
if( SUCCEEDED( hResult ) ){
if( ( pPhrase->pProperties != nullptr ) && ( pPhrase->pProperties->pFirstChild != nullptr ) ){
// 辞書のフレーズタグと比較
const SPPHRASEPROPERTY* pSemantic = pPhrase->pProperties->pFirstChild;
if( pSemantic->SREngineConfidence > confidenceThreshold ){
if( wcscmp( L"あか", pSemantic->pszValue ) == 0 ){
std::cout << "あか" << std::endl;
audioMat = cv::Scalar( 0, 0, 255 );
}
else if( wcscmp( L"みどり", pSemantic->pszValue ) == 0 ){
std::cout << "みどり" << std::endl;
audioMat = cv::Scalar( 0, 255, 0 );
}
else if( wcscmp( L"あお", pSemantic->pszValue ) == 0 ){
std::cout << "あお" << std::endl;
audioMat = cv::Scalar( 255, 0, 0 );
}
else if( wcscmp( L"おわり", pSemantic->pszValue ) == 0 ){
exit = true;
}
}
}
CoTaskMemFree( pPhrase );
}
}
break;
default:
break;
}
pSpeechContext->GetEvents( 1, &eventStatus, &eventFetch );
}
}
// 表示
cv::imshow( "Audio", audioMat );
// ループの終了判定(Escキー)
if( cv::waitKey( 30 ) == VK_ESCAPE || exit ){
break;
}
}
// 終了処理
audioStream->StopCapture();
pSpeechRecognizer->SetRecoState( SPRST_INACTIVE );
CoUninitialize();
pSensor->NuiShutdown();
CloseHandle( hSpeechEvent );
cv::destroyAllWindows();
return 0;
}
