none
Australian recognition difficulty RRS feed

  • Question

  • Hi all,

    I've just been playing so far with the audio features and I'm mostly fairly impressed.

    It took me a little while before I realised that everything was hypothesizing and selecting blue because IT WAS THE LAST ONE IN THE LIST. I added another random word to the end of the list then every non-matching word became that in stead. When the confidence model actually works, that won't be an issue but sure was a bit of a gotcha in my initial playing.

    Anyway, it's suddenly in my bad books because my throat is now sore from bunging on an American accent to get it to recognise numbers properly. I simply replaced the colour words in the Speech sample with digits;

     

                        var numbers = new Choices();
                        numbers.Add("1");
                        numbers.Add("2");
                        numbers.Add("3");
                        numbers.Add("4");
                        numbers.Add("5");
                        numbers.Add("6");
                        numbers.Add("7");
                        numbers.Add("8");
                        numbers.Add("9");
                        numbers.Add("0");
    

     


    Then into the GrammarBuilder, Grammaar, and SpeechRecognitionEngine's .LoadGrammar() etc.

    I managed to get "8" across eventually, if I tense up totally. I managed one or two "4"s but many more attempts and I'm going to have some kind of hernia... Every time I just say "4" normally (in my Australian accent) I get "7". I also get "7" when I say "7", so that's a plus. The other digits worked fine.

    I tried changing the predicate in the GetKinectRecognizer() method to Australia but get a null result.

     

    private static RecognizerInfo GetKinectRecognizer()
            {
                Func<RecognizerInfo, bool> matchingFunc = r =>
                {
                    string value;
                    r.AdditionalInfo.TryGetValue("Kinect", out value);
                    return "True".Equals(value, StringComparison.InvariantCultureIgnoreCase) && "en-AU".Equals(r.Culture.Name, StringComparison.InvariantCultureIgnoreCase);
                };
                return SpeechRecognitionEngine.InstalledRecognizers().Where(matchingFunc).FirstOrDefault();
            }
    
    

     

    Am I doing something stupid, or has Australian been left out of the Kinect recognizers?

    Additional: I changed the list to "one", "two", "three", "four", "five" etc and showed the confidence in the output. "eight" only gets about 45% confidence unless I slather on the American accent, "four" works just fine, but now "two" keeps coming up with "seven". How do you guys pronounce "seven"? Or is it that I'm making everything sound like it? :p I would've thought the two syllables would've made it last choice though...

    Thanks

    Jeff



    • Edited by Jeff Whitty Monday, December 12, 2011 11:42 AM
    Monday, December 12, 2011 11:35 AM

Answers

  • We just announced that we will be adding a number of languages and locales in our 1.5 release at the end of May, including en-AU. :)

    We are also hard at work on our 1.5 release, which will be available at
    the end of May.  ...
    Also included in our 1.5 release will be four new languages for speech
    recognition – French, Spanish, Italian, and Japanese.  In addition, we
    will be releasing new language packs which enable speech recognition for
    the way a language is spoken in different regions: English/Great
    Britain, English/Ireland, English/Australia, English/New Zealand,
    English/Canada, French/France, French/Canada, Italian/Italy,
    Japanese/Japan, Spanish/Spain and Spanish/Mexico.

    http://blogs.msdn.com/b/kinectforwindows/archive/2012/03/26/what-s-ahead-a-sneak-peek.aspx

    If you need to do other locales before that you can use the SAPI engine, but the recognition will not be as good, as it was not trained w/ the Kinect input.

    Tuesday, March 27, 2012 3:53 PM

All replies

  • If I remember correctly the only Kinect language out right now is US English. If I'm wrong I'm sure they'll correct me, but otherwise I wouldn't expect that to change much until this SDK goes commercial. 
    Wednesday, December 14, 2011 2:03 AM
  • I think if you program straight through SAPI it will allow you to access different language models? For us Aussies, I would suggest en-GB instead of en-US; it's much closer.

    Clever, witty forum signature.
    Monday, January 30, 2012 11:09 PM
  • We just announced that we will be adding a number of languages and locales in our 1.5 release at the end of May, including en-AU. :)

    We are also hard at work on our 1.5 release, which will be available at
    the end of May.  ...
    Also included in our 1.5 release will be four new languages for speech
    recognition – French, Spanish, Italian, and Japanese.  In addition, we
    will be releasing new language packs which enable speech recognition for
    the way a language is spoken in different regions: English/Great
    Britain, English/Ireland, English/Australia, English/New Zealand,
    English/Canada, French/France, French/Canada, Italian/Italy,
    Japanese/Japan, Spanish/Spain and Spanish/Mexico.

    http://blogs.msdn.com/b/kinectforwindows/archive/2012/03/26/what-s-ahead-a-sneak-peek.aspx

    If you need to do other locales before that you can use the SAPI engine, but the recognition will not be as good, as it was not trained w/ the Kinect input.

    Tuesday, March 27, 2012 3:53 PM