Speech Recognition Engine returning grammar found from wrong speech

Question

I'm trying to use the .NET SpeechRecognitionEngine with C# in VisualStudio Express. However i'm finding that it's picking up completely wrong words / sentences and assuming they are something in the grammar.

EG If I load "test 1" into the grammar and say "filthy beast" which is not even close to the words "test 1", the EventHandler SpeechRecognized fires. I left a movie playing on netflix while coding and it was firing the recognized event to music and talk in the movie, so it's way way off.

Is there a way to prevent it from assuming the spoken words are in the grammar? Or any way to stop this?

Any tips?

Here is a log output for me saying "filthy beast" when the grammar only has "test 1" loaded into it.

speechDetectedHandler():
speechHypothesizedHandler():  confidence = 0.002903746    e.Result.Text = Test
speechHypothesizedHandler():  confidence = 0.8096436    e.Result.Text = Test
speechRecognizedHandler():  confidence = 0.7723699    e.Result.Text = Test 1

Code:

public SpeechRecognitionEngine sre;

String culture = "en-US";
foreach (RecognizerInfo config in SpeechRecognitionEngine.InstalledRecognizers())
{
    if (config.Culture.ToString() == culture)
    {
      s = new SpeechRecognitionEngine(config);
      break;
    }
}
s.SetInputToDefaultAudioDevice();

sre.MaxAlternates = 0;

sre.AudioLevelUpdated += new EventHandler<AudioLevelUpdatedEventArgs>(audioLevelHandler);
sre.SpeechRecognized += new EventHandler<SpeechRecognizedEventArgs>(speechRecognizedHandler);

sre.SpeechHypothesized += new EventHandler<SpeechHypothesizedEventArgs>(speechHypothesizedHandler);
sre.SpeechDetected += new EventHandler<SpeechDetectedEventArgs>(speechDetectedHandler);


gb = new GrammarBuilder(speechCommands);
g = new Grammar(gb);

sre.UnloadAllGrammars();
sre.LoadGrammar(g);
startListening();

I have edited your title. Please see, "Should questions include “tags” in their titles?", where the consensus is "no, they should not". — John Saunders

Darcey Darcey · Accepted Answer · 2015-05-07T21:08:19

Solution is to create and load grammar that is similar to the word / grammar / speech you want to use, this will increase accuracy. Then to evaluate hypothesized trigger 1, trigger 2 and then recognized confidence levels and result text. Not very practical as this would be different for each person / user.

There is no way to prevent the .NET Speech Recognition Engine from ALWAYS RETURNING A GRAMMAR MATCH. You may as well be saying "bob" in a silent room into a studio grade mic and it would recognize "open windows media player". lol

Warning 1: grammar word lists of over 1,000 slow things down and can lock the application.

Warning 2: en-US has good english recognition capabilities, switching to en-GB etc lowers accuracy drastically

So far with Googles Speech Recognition API (does require you to be online) but it is 10x more accurate and you can easily test for a match yourself.

Speech Recognition Engine returning grammar found from wrong speech

2 Answers