void SpeechRecognition::InitContext()
{
//init com
if (FAILED(::CoInitialize(nullptr))) {
Log("Failed to initialize com library", LogEntry::Error);
return;
}
HRESULT hr;
ISpRecognizer* recognizer;
hr = CoCreateInstance(CLSID_SpSharedRecognizer, nullptr, CLSCTX_ALL, IID_ISpRecognizer, reinterpret_cast<void**>(&recognizer));
CheckReturn(hr);
hr = recognizer->CreateRecoContext(&recoContext);
CheckReturn(hr);
//pause context
hr = recoContext->Pause(0);
CheckReturn(hr);
//make grammar library
ISpRecoGrammar* recoGrammar = InitGrammar();
handleEvent = recoContext->GetNotifyEventHandle();
if (handleEvent == INVALID_HANDLE_VALUE)
CheckReturn(E_FAIL);
ULONGLONG interest = SPFEI(SPEI_RECOGNITION);
hr = recoContext->SetInterest(interest, interest);
CheckReturn(hr);
//Activate grammar
hr = recoGrammar->SetRuleState(RULENAME, 0, SPRS_ACTIVE);
CheckReturn(hr);
//enable context again
hr = recoContext->Resume(0);
CheckReturn(hr);
std::cout << "Waiting for mocha...." << std::endl;
Update();
std::cout << "Hello!" << std::endl;
}
ISpRecoGrammar* SpeechRecognition::InitGrammar()
{
HRESULT hr;
SPSTATEHANDLE sate;
ISpRecoGrammar* recoGrammar;
hr = recoContext->CreateGrammar(GRAMMARID, &recoGrammar);
CheckReturn(hr);
WORD langID = MAKELANGID(LANG_ENGLISH, SUBLANG_ENGLISH_US);
hr = recoGrammar->ResetGrammar(langID);
CheckReturn(hr);
//create rules
hr = recoGrammar->GetRule(RULENAME, 0, SPRAF_TopLevel | SPRAF_Active, true, &sate);
CheckReturn(hr);
//add words
const std::wstring commandWstr = std::wstring(L"Mocha");
hr = recoGrammar->AddWordTransition(sate, NULL, commandWstr.c_str(), L" ", SPWT_LEXICAL, 1, nullptr);
CheckReturn(hr);
for (std::wstring& ws : wordList)
{
std::string sentence(ws.begin(), ws.end());
std::vector<std::string> words = split(sentence, ' ');
for (std::string& w : words)
{
std::wstring wsword(w.begin(), w.end());
hr = recoGrammar->AddWordTransition(sate, NULL, wsword.c_str(), L" ", SPWT_LEXICAL, 1, nullptr);
}
CheckReturn(hr);
}
hr = recoGrammar->Commit(0);
CheckReturn(hr);
return recoGrammar;
}
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;
}