Microsoft SAPI Sub-language issue

Question

My problem is this: My SAPI inproc recognizer refuses to load my grammar file when the windows 10 Display language is set to English(UK).

System Display language set to UK. Speech Recognition language is UK. System Locale is UK. SAPI xml-format Grammar even specifies LANGID=809 - as far as I can tell, EVERYTHING is set to EN-GB, and yet the grammar still won't load.

But it loads and works just fine when display language and is set to English(US).

Does anyone know what's up with this? It's extremely frustrating... Hopefully I'm just missing something simple.

SAPI Initialization code:

    //////////////INITIALIZE SAPI ENGINE AND GRAMMAR//////////////////////////////
HRESULT SpeechObject::Initialize(){
    //INITIALIZE SR ENGINE
    if (FAILED(test=::CoInitialize(NULL)))
        SRError(L"COM Initialization Fail");

    //Create recognizer instance
    if (FAILED(test=cpEngine.CoCreateInstance(CLSID_SpInprocRecognizer))){
        SRError(L"Can't Load Reco Engine");
            return test;
    }

    //Load the audio Input (in seperate function to facilitate reload)
    LoadAudio(); //should I check this?

    //load Default recognizer settings
    cpEngine->SetRecognizer(NULL);

    //get and load default reco profile
    if (FAILED(SpGetDefaultTokenFromCategoryId(SPCAT_RECOPROFILES, &cpObjectToken)))
        SRError(L"Can't Find Recognition Profile");
    if (FAILED(cpEngine->SetRecoProfile(cpObjectToken)))
        SRError(L"Can't Load Recognition Profile");

    //create reco context
    if (FAILED(test=cpEngine->CreateRecoContext(&cpContext))){
        SRError(L"Can't Create Reco Context");
        return test;
    }

    //send pSpeechObject to global callback function
    cpContext->SetNotifyCallbackFunction(
        (SPNOTIFYCALLBACK*)SpeechCallBack,
        NULL, (LPARAM)this);

    if(FAILED(cpContext->CreateGrammar(NULL, &cpGrammar)))
        SRError(L"Can't Create context");

    char str[80]; ////TEST
    sprintf(str, "LANGID: %X", GetUserDefaultUILanguage());
    MessageBoxA(GetActiveWindow(), str,0,0);

    //load grammar from compiled grammar resource
    if (FAILED(test = cpGrammar->LoadCmdFromResource(
        hModule, MAKEINTRESOURCE(GRAMMARCFG),
        L"FILE", GetUserDefaultUILanguage(), SPLO_STATIC))){
        SRError(L"Can't Load Grammar. Please check language settings");
        return test;
    }

    //(comment above and uncomment following to load from raw xml file for testing)
    //cpGrammar->LoadCmdFromFile(L"Grammar.xml", SPLO_STATIC);

    //Enable Engine and Reco Context
    cpEngine->SetRecoState(SPRST_ACTIVE);
    cpContext->SetContextState(SPCS_ENABLED);

    //enable ALWAYS ACTIVE and GROUND ENGINES ON commands
    return(cpGrammar->SetRuleState(NULL, NULL, SPRS_ACTIVE));

}

////////////LOAD (AND RELOAD) AUDIO INPUT//////////////////////
HRESULT SpeechObject::LoadAudio(bool dlgFlag){
    if (FAILED(test = SpCreateDefaultObjectFromCategoryId(SPCAT_AUDIOIN, &cpAudioIn))){
        SRError(L"Can't Find Default Audio Input");
        return test;
    }

    if (FAILED(test = cpEngine->SetInput(cpAudioIn, TRUE))){
        if (!dlgFlag)
            SRError(L"Can't Set Audio Input");
        return test;
    }

    if (pSRDisplay)
        pSRDisplay->DisplayText("Audio Reloaded");
    if (pDLog)
        pDLog->LogEvent("Audio Reloaded");
    //RecoState must be reenabled after audio reset
    cpEngine->SetRecoState(SPRST_ACTIVE);
    if (pDLog)
        pDLog->LogEvent("SR ENABLED");
    return test;
}

I get the "Can't Load Grammar. Please check language settings" error any time the display language is not English(US), even if I confirm that ALL SETTINGS match...

Would really appreciate any sort of insight from persons more knowledgeable than I.

Farley

score 0 · Answer 1 · answered Dec 04 '15 at 00:30

0

You need to explicitly load the recognizer for your preferred language. In particular, this:

//load Default recognizer settings
cpEngine->SetRecognizer(NULL);

always loads the recognizer specified in the speech control panel. You likely want something like this:

CComPtr<ISpObjectToken> cpEngineToken;
hr = SpFindBestToken(SPCAT_RECOGNIZERS, L"Language=<hex language id>", NULL, &cpEngineToken);
// check hr
hr = cpEngine->SetRecognizer(cpEngineToken);

where you'll need to convert the LCID from GetUserDefaultUILanguage to a hex number.

answered Dec 04 '15 at 00:30

Eric Brown

13,774
7
30
71

Unfortunately, this doesn't seem to work. Indeed, I would by quite surprised if it did, considering the setting in the speech control panel already matched the display language. All the settings matched. I really appreciate your help, though, and I am eager to hear any other ideas! – lequinne Dec 04 '15 at 02:12
works with display language 409 (US), but not with 809 (UK). Neither SpFindBestToken() nor SetRecognizer() fail. In other words, I still have exactly the same problem. – lequinne Dec 04 '15 at 02:28
Do you, in fact, have a recognizer engine installed for en-UK? My Win10 machine, for example, does not. – Eric Brown Dec 04 '15 at 17:18
Yes sir. I installed it myself, and its existence is underscored by the fact that when I try passing different language codes to SpFindBestToken(), all calls return "failed" except "Language=409" and "Language=809", the two engines I have installed. – lequinne Dec 04 '15 at 18:25
With an inproc recognizer, the settings in the speech control panel are more or less irrelevant. You're responsible for setting everything up. At this point, I'd switch to LoadGrammarFromFile or LoadGrammarFromMemory, so that you can separate out "loading the grammar at all" from "loading the grammar out of the resource file" - because the langid for the LoadCmdFromResource file is the langid of the resource file, not the langid of the grammar. – Eric Brown Dec 04 '15 at 20:26
Also, I'd remove the calls to GetUserDefaultUILanguage() and explicitly specify 0x809 everywhere (except for LoadCmdFromResource()) to track down the failures. – Eric Brown Dec 04 '15 at 20:37

score 0 · Accepted Answer · answered Dec 05 '15 at 03:32

The issue was that I had misinterpreted the meaning of the "language" parameter in LoadCmdFromResource(). I'll blame it on the ambiguous SAPI documentation, though if I had experience loading some other types of resources before I might have been tipped off to this. ;) I had thought it was somehow used by SAPI and should match the language of the system and recognizer (that's what it sounded like in the documentation). Turns out, it actually just specifies the language used to compile the .RC file the grammar is included in (presumably to allow multiple translations to be included in separate .rc's).

The code works perfectly as originally posted, so long as I replace "GetUserDefaultUI()" with an explicit "0x409" (the language specified in the resource compiler) in the call to LoadCmdFromResource(). Now it works with US English, UK English and presumably all English recognizers, and loads the recognizer selected in the speech control panel regardless of the Display language setting (which can even be non-English).

Many, many thanks to Eric Brown for tipping me off to this, I was starting to lose my mind.

Farley

Microsoft SAPI Sub-language issue

2 Answers2