none
Having trouble with adding speech reco to another sample RRS feed

  • Question

  • Hi, I'm having a similar problem... I added KinectAudioStream code from speech basics d2d to the face tracking visualization... Add initialization for audio stream, and recognizer to the initialization part, and create another thread to process the speech... when i run the program, most of the time the speech thread doesn't recognize anything, sometimes it recognize something with a very low confidence value and sometimes it mistakenly recognize something with a very high confidence value although i wasn't talking anything. I suspect it has something to do with the audio stream, since when i tried to use a prerecorded wav file, the recognizer always able to recognize the words with high accuracy. Can anyone help? Thanks.

    *update

    I've tried the solution for calling process speech after WaitForMultipleObjects(...) but it's not helping much... it recognize some words but with a very low confidence level 0.002. anything wrong with my audio stream initialization?

    Friday, June 8, 2012 9:46 AM

All replies

  • A couple of quick questions:

    1) Is speech basics working well for you by itself?

    2) What do you mean "when you use a prerecorded wave file"? 

    Friday, June 8, 2012 3:27 PM
  • 1) yes, the speech basics work well by itself

    2) I load a wav file and use it as a stream to the speech recognizer using ispstream::bindtofile

    Sunday, June 10, 2012 3:14 AM