SAPI: Speech to Text example

Question

I am new to SAPI, and I would really appreciate if any of you can provide me a speech to text Hello World example in SAPI. I know MS got some examples like "Dictation" etc, but I would like to start with a very small one. Glad if you can help.

Tyika

Answer 1

I played a bit with Windows Voice Recognition using SAPI, it really isn't user friendly. Here is an example of code I wrote (in C++) :

#include <sphelper.h>
#include <sapi.h>
#include <iostream>
#include <string>

const ULONGLONG grammarId = 0;
const wchar_t* ruleName1 = L"ruleName1";

int start_listening(const std::string& word);
ISpRecoGrammar* init_grammar(ISpRecoContext* recoContext, const std::string& command);
void get_text(ISpRecoContext* reco_context);
void check_result(const HRESULT& result);

int main(int argc, char** argv)
{
    start_listening("Hello");
    return EXIT_SUCCESS;
}

// This function exits when the word passed as parameter is said by the user
int start_listening(const std::string& word)
{
    // Initialize COM library
    if (FAILED(::CoInitialize(nullptr))) {
        return EXIT_FAILURE;
    }

    std::cout << "You should start Windows Recognition" << std::endl;
    std::cout << "Just say \""<< word << "\"" << std::endl;

    HRESULT hr;

    ISpRecognizer* recognizer;
    hr = CoCreateInstance(CLSID_SpSharedRecognizer,
        nullptr, CLSCTX_ALL, IID_ISpRecognizer,
        reinterpret_cast<void**>(&recognizer));
    check_result(hr);

    ISpRecoContext* recoContext;
    hr = recognizer->CreateRecoContext(&recoContext);
    check_result(hr);

    // Disable context
    hr = recoContext->Pause(0);
    check_result(hr);

    ISpRecoGrammar* recoGrammar = init_grammar(recoContext, word);

    hr = recoContext->SetNotifyWin32Event();
    check_result(hr);

    HANDLE handleEvent;
    handleEvent = recoContext->GetNotifyEventHandle();
    if(handleEvent == INVALID_HANDLE_VALUE) {
        check_result(E_FAIL);
    }

    ULONGLONG interest;
    interest = SPFEI(SPEI_RECOGNITION);
    hr = recoContext->SetInterest(interest, interest);
    check_result(hr);

    // Activate Grammar
    hr = recoGrammar->SetRuleState(ruleName1, 0, SPRS_ACTIVE);
    check_result(hr);

    // Enable context
    hr = recoContext->Resume(0);
    check_result(hr);

    // Wait for reco
    HANDLE handles[1];
    handles[0] = handleEvent;
    WaitForMultipleObjects(1, handles, FALSE, INFINITE);
    get_text(recoContext);

    std::cout << "Hello user" << std::endl;

    recoGrammar->Release();
    ::CoUninitialize();

    system("PAUSE");
    return EXIT_SUCCESS;
}

/**
* Create and initialize the Grammar.
* Create a rule for the grammar.
* Add word to the grammar.
*/
ISpRecoGrammar* init_grammar(ISpRecoContext* recoContext, const std::string& command)
{
    HRESULT hr;
    SPSTATEHANDLE sate;

    ISpRecoGrammar* recoGrammar;
    hr = recoContext->CreateGrammar(grammarId, &recoGrammar);
    check_result(hr);

    WORD langId = MAKELANGID(LANG_FRENCH, SUBLANG_FRENCH);
    hr = recoGrammar->ResetGrammar(langId);
    check_result(hr);
    // TODO: Catch error and use default langId => GetUserDefaultUILanguage()

    // Create rules
    hr = recoGrammar->GetRule(ruleName1, 0, SPRAF_TopLevel | SPRAF_Active, true, &sate);
    check_result(hr);

    // Add a word
    const std::wstring commandWstr = std::wstring(command.begin(), command.end());
    hr = recoGrammar->AddWordTransition(sate, NULL, commandWstr.c_str(), L" ", SPWT_LEXICAL, 1, nullptr);
    check_result(hr);

    // Commit changes
    hr = recoGrammar->Commit(0);
    check_result(hr);

    return recoGrammar;
}

void get_text(ISpRecoContext* reco_context)
{
    const ULONG maxEvents = 10;
    SPEVENT events[maxEvents];

    ULONG eventCount;
    HRESULT hr;
    hr = reco_context->GetEvents(maxEvents, events, &eventCount);

    // Warning hr equal S_FALSE if everything is OK 
    // but eventCount < requestedEventCount
    if(!(hr == S_OK || hr == S_FALSE)) {
        check_result(hr);
    }

    ISpRecoResult* recoResult;
    recoResult = reinterpret_cast<ISpRecoResult*>(events[0].lParam);

    wchar_t* text;
    hr = recoResult->GetText(SP_GETWHOLEPHRASE, SP_GETWHOLEPHRASE, FALSE, &text, NULL);
    check_result(hr);

    CoTaskMemFree(text);
}

void check_result(const HRESULT& result)
{
    if (result == S_OK) {
        return;
    }

    std::string message;

    switch(result) {

    case E_INVALIDARG:
        message = "One or more arguments are invalids.";

    case E_ACCESSDENIED:
        message = "Acces Denied.";

    case E_NOINTERFACE:
        message = "Interface does not exist.";

    case E_NOTIMPL:
        message = "Not implemented method.";

    case E_OUTOFMEMORY:
        message = "Out of memory.";

    case E_POINTER:
        message = "Invalid pointer.";

    case E_UNEXPECTED:
        message = "Unexpecter error.";

    case E_FAIL:
        message = "Failure";

    default:
        message = "Unknown : " + std::to_string(result);
    }

    throw std::exception(message.c_str());
}

As I said, it's a bit complicated. I think you should wrap all that code into a library to make it easier to use.

Subvene

Yes, I am moving with C++/CLI. So I can use simple C# code :) – Tyika

In front of the awful C++ API I thought of going with C# too. But I like C++ too much ;-) – Subvene

Just want to make you all aware certain parts of the. Net implementation doesn't work. Also when you get into advanced stuff some bits are not supported in the .NET api. – Monasticism

Compared to other APIs that does the same, this is a horrible API. Thank you for the example. – Castorena