c++ - 训练 sapi : Creating transcripted wav files and adding file paths to registry

Question

我们正在尝试进行声学训练，但无法创建转录的音频文件，如何创建？此外，我们正在使用 GetTranscript 和 Appendtranscript，但如果我们以 READWRITE 模式打开流，则无法获取 ISpStream 的 ISpTranscript 接口(interface)，那么如何创建转录 wav 文件。

hr = SPBindToFile(L"e:\\file1.wav", SPFM_OPEN_READONLY,
    &cpStream);
hr = cpStream.QueryInterface(&cpTranscript);
// We get a error here for as E_NONINTERFACE if SPFM_OPEN_READWRITE  
hr = cpTranscript->AppendTranscript(sCorrectText);
hr = cpTranscript->GetTranscript(&pwszTranscript);
// GIVES CORRECT TRANSCRIPT 

//READING THIS AGAIN ON NEXT EXECUTION TIME DOES NOT GIVE THE TRANSCRIPT

hr = SPBindToFile(L"e:\\file1.wav", SPFM_OPEN_READONLY,
    &cpStream);
hr = cpStream.QueryInterface(&cpTranscript);
//THIS GIVE THE ERROR E_NONINTERFACE

完成此操作后，我们需要将文件路径添加到注册表中。我们通过以下代码执行此操作。

CComPtr<ISpObjectToken> cpObjToken;



ULONG                     CSIDL_LOCAL_APPDATA = 28;
ULONG                     CSIDL_FLAG_CREATE = 32768;
GUID guid0;
LPWSTR FileName2;

hr = cpRecognizerBase->GetRecoProfile(&cpObjToken);
hr = CoCreateGuid(&guid0);
hr = cpObjToken->GetStorageFileName(guid0, L"Test",   L"F:\\sample6.wav",CSIDL_FLAG_CREATE, &FileName2);
//this code runs fine but the file is never added to the registry

任何指针将不胜感激。此问题与此处提出的问题有关 Speech training files and registry locations

谢谢

Answer 1

In this post I address how to perform appendTranscript successfully, and speech training using WAV files (credit to Bill Hutchinson). Everything is in C++.

如果 ISPStream 没有内容，就会发生 E_NONINTERFACE。例如文件是空的；调用没有成功，但仍然返回 s_OK(它出于某种原因这样做)。所以通常我会先调查流是否真的有任何内容。您可以通过检查其大小来做到这一点:

这是一个例子。如果它的大小为 0 或大得离谱，那么显然它没有返回正确的值。请记住，返回值是 ULARGE_INTEGER。

STATSTG streamInfo;
cpStream->Stat(&streamInfo, STATFLAG_DEFAULT);
ULARGE_INTEGER streamSizeULI;
streamSizeULI = streamInfo.cbSize;

SPBindToFile 仅适用于 SPFM_OPEN_READONLY 和 SPFM_CREATE_ALWAYS，因此您必须使用其中之一。

至于如何保存附加的转录本，如果wav文件已经存在(或者至少我不知道如何保存)，似乎不能直接保存。如果该文件尚不存在，您可以创建一个新的 ispstream，当您通过语音或麦克风等方式将音频信息传递给它时(网络上有很多示例)，您可以附加一个文字记录，然后它会粘住.我在下面举了一个例子。

将抄本附加到新文件:

void recordAndAppendTranscriptInOneOperation() {
HRESULT             hr = S_OK;
CComPtr <ISpVoice>      cpVoice;
CComPtr <ISpStream>     cpStream;
CComPtr<ISpTranscript>  cpTranscript;
CSpStreamFormat         cAudioFmt;

//Create    a   SAPI    Voice   
hr  =   cpVoice.CoCreateInstance(CLSID_SpVoice);

char filePathOut[] = R"(C:\SAPI\SampleOutput\SP_Sample.wav)";

//Set   the audio   format                              
if(SUCCEEDED(hr))   
{       
    hr  =   cAudioFmt.AssignFormat(SPSF_22kHz16BitMono);    
}

//Call  SPBindToFile,   a   SAPI    helper  method,     to  bind    the audio       
if(SUCCEEDED(hr))   
{
    hr = SPBindToFile(filePathOut, SPFM_CREATE_ALWAYS, &cpStream, &cAudioFmt.FormatId(), cAudioFmt.WaveFormatExPtr());
}

//set   the output  to  cpStream    so  that    the output  audio   data    wil                             
if(SUCCEEDED(hr))   
{       
    hr = cpVoice->SetOutput(cpStream, TRUE);    
}

//Speak the text    “hello  world”  synchronously                               
if(SUCCEEDED(hr))   
{       
    hr = cpVoice->Speak(L"Hello World", SPF_DEFAULT, NULL);
}

//close the stream  
if(SUCCEEDED(hr))   
{
    PWCHAR                      pwszTranscript;
    char NewTranscriptAsString[] = R"(This is a test)";
    LPCWSTR NewTranscript = charToLPSTRW(NewTranscriptAsString);

    hr = cpStream.QueryInterface(&cpTranscript);
    hr = cpTranscript->AppendTranscript(NULL);
    hr = cpTranscript->AppendTranscript(NewTranscript);
    hr = cpTranscript->GetTranscript(&pwszTranscript);

    hr  =   cpStream->Close();  
}

//Release   the stream  and voice   object  
cpStream.Release(); 
cpVoice.Release();
 }

Bill Hutchinson(下面的链接来源之一)有一些代码可用于执行识别器训练，而无需进行所有注册表编辑等操作。我把它包含在这篇文章的末尾。他有一个函数(TrainOne)，可以通过内存流逐个文件地训练识别器。您可以将预先存在的 WAV 传递给它。具体来说，要么是带有转录本的 WAV，要么是没有转录本的 WAV，并且(然后在调用时将转录本提供给函数)。请查看它，因为它提供了很多信息。

这里是我发现的与 SAPI 相关的所有知识的集合，这对其他试图解决这个问题的人很有用。 我也会很快发布我自己的完整 SAPI 培训解决方案:

• How to use the function GetStorageFileName for adding training files to registry?

• Acoustic training using SAPI 5.3 Speech API

• Training sapi : Creating transcripted wav files and adding file paths to registry

• https://groups.google.com/forum/#!topic/microsoft.public.speech_tech.sdk/fTq-PJrVd_Q

• https://documentation.help/SAPI-5/documentation.pdf

示例训练代码:

由于 Bill Hutchinson 的 SAPI 代码是如何使用 SAPI 在网络上进行培训的少数可靠示例之一，我在下面包含了他来自谷歌的帖子，以防有一天它被删除/丢失:

#include "stdafx.h"
#include "sphelper.h"
#include <sapi.h>
#include <string.h>
//MAIN() is last function below
inline HRESULT ReturnResult(ISpRecoContext * pRecoCtxt, ISpRecoResult
** ppResult)
{
        HRESULT hr = S_OK;
        CSpEvent spEvent;
        while (S_OK == pRecoCtxt->WaitForNotifyEvent(INFINITE))
        {
                while (S_OK == spEvent.GetFrom(pRecoCtxt))
                {
                        switch (spEvent.eEventId)
                        {
                                case SPEI_RECOGNITION:
                                        *ppResult = spEvent.RecoResult();
                                        if (*ppResult)
                                        {
                                                (*ppResult)->AddRef();
                                        }
                                        return hr;
                                case [OTHER EVENTS]
                    spEvent.Clear();
        }
        return hr;
}
inline HRESULT TrainOneFile(ISpRecoContext * cpRecoCtxt, ISpRecognizer
* cpRecognizerBase, ISpRecoGrammar * cpGrammar)
{
        CComPtr<ISpStream>      cpStream;
        CComPtr<ISpRecoResult>        cpResult;
        CComPtr<ISpTranscript>  cpTranscript;
        PWCHAR                  pwszTranscript;
        HRESULT hr = S_OK;
        hr = cpStream.CoCreateInstance(CLSID_SpStream);
        // Bind a stream to an existing wavefile
        if (SUCCEEDED(hr))        {
                hr = cpStream->BindToFile(L"C:\\XX.wav",                                                        SPFM_OPEN_READONLY,
                        NULL,
                        NULL,
                        SPFEI_ALL_EVENTS);
        }
        if (SUCCEEDED(hr)){
                hr = cpStream.QueryInterface(&cpTranscript);
        }
        if (SUCCEEDED(hr)) {
                hr = cpTranscript->GetTranscript(&pwszTranscript);
        }
        //THIS IS ALTERNATE CODE FOR PREVIOUS LINE, FOR SOUND FILES THAT
DON’T HAVE A TRANSCRIPT ATTACHED
        LPCWSTR sCorrectText = L"Anyone who has spent time on a farm knows
there is a rhythm to the year.";
        if (SUCCEEDED(hr)){
                hr = cpTranscript->AppendTranscript(s);
        }
        if (SUCCEEDED(hr))        {
                hr = cpTranscript->GetTranscript(&pwszTranscript);
        }
        if(SUCCEEDED(hr)){
                hr = cpRecognizerBase->SetInput(cpStream, TRUE);
        }
        USES_CONVERSION;
        CSpDynamicString dstrText;
        if (SUCCEEDED (hr)){
                hr = cpGrammar->SetDictationState(SPRS_ACTIVE);
        }
        if (SUCCEEDED(hr)){
                hr = ReturnResult(cpRecoCtxt, &cpResult);
        }
        if (SUCCEEDED(hr)){
                hr = cpGrammar->SetDictationState( SPRS_INACTIVE );
        }
        if ((cpResult) &&(SUCCEEDED(hr))){
                hr = cpResult-
>GetText(SP_GETWHOLEPHRASE,SP_GETWHOLEPHRASE,TRUE,&dstrText,NULL);
        }
        CComPtr<ISpRecoResult2> cpResult2;
        if (SUCCEEDED(hr)){
                hr = cpResult.QueryInterface<ISpRecoResult2>(&cpResult2);
        }
        if (SUCCEEDED(hr)){
//COMMITTEXT SHOULD FORCE ADAPTATION OF MODELS TO CORRECT TEXT
//(THO IT SHOULD BE REDUNDANT WITH SETTRAININGSTATE() ?)
                hr = cpResult2-
>CommitText(SP_GETWHOLEPHRASE,SP_GETWHOLEPHRASE,sCorrectText,SPCF_DEFINITE_CORRECTION);
                cpResult.Release();
                cpResult2.Release();
        }
        return hr;
}

int _tmain(int argc, _TCHAR* argv[])
{
        HRESULT hr = S_OK;
        CComPtr<ISpRecognizer2> cpRecognizer;
        CComPtr<ISpRecoContext> cpRecoCtxt;
        CComPtr<ISpRecoGrammar> cpGrammar;
        CComPtr<ISpRecognizer> cpRecognizerBase;
        hr = ::CoInitialize(NULL);
            if (SUCCEEDED(hr)) {
                hr = cpRecognizer.CoCreateInstance(CLSID_SpInprocRecognizer);
        }
        if (SUCCEEDED(hr)){
                hr = cpRecognizer.QueryInterface<ISpRecognizer>(&cpRecognizerBase);
        }
        if (SUCCEEDED(hr)){
                hr = cpRecognizerBase->CreateRecoContext(&cpRecoCtxt);
        }
        if (cpRecoCtxt){
                hr = cpRecoCtxt->CreateGrammar(0, &cpGrammar);
        }
        if (SUCCEEDED(hr)){
                hr = cpGrammar->LoadDictation(NULL, SPLO_STATIC);
        }
        if (SUCCEEDED(hr)){
                hr = cpRecognizer->SetTrainingState(TRUE, TRUE);
        }
        if (SUCCEEDED(hr)){
                hr = cpRecoCtxt->SetNotifyWin32Event();
        }
        if (SUCCEEDED(hr)){
                hr = cpRecoCtxt->SetInterest(
                        SPFEI(SPEI_RECOGNITION)|
                        SPFEI(SPEI_HYPOTHESIS)|
                        SPFEI(SPEI_FALSE_RECOGNITION),
                        SPFEI(SPEI_RECOGNITION)|
                        SPFEI(SPEI_HYPOTHESIS)|
                        SPFEI(SPEI_FALSE_RECOGNITION));
        }
        if (SUCCEEDED(hr)){
                hr = TrainOneFile(cpRecoCtxt, cpRecognizerBase, cpGrammar);
        }
        if (SUCCEEDED(hr)){//RERUN TO CHECK FOR IMPROVEMENT
                hr = TrainOneFile(cpRecoCtxt, cpRecognizerBase, cpGrammar);
        }
        cpRecognizer->SetTrainingState(FALSE, TRUE);//should turn off and
save changes
        ::CoUninitialize();
        return 0;
}