Speech Recognition Performance RRS feed

  • Question

  • Hi - I noticed that the speech recognition performance is between 3.5 and 8 milliseconds to achieve a recognition result, but it performance seems uneven and it takes too long to achieve a recognized result for my needs.  Is there a way to be able to tune how quickly it takes the engine to achieve a recognition result - even at the expense of the quality of recognized speech?  Currently using the v1.x sdk, but open to upgrading to v2.x if required.



    Tuesday, April 14, 2015 2:47 AM

All replies

  • As far as I know, there is a confidence value that you must set in voice recognition. This is sort of a threshold if the user's speech does not hit this value, Kinect will not accept it. So if you set this value to be very low (0.2 for example), Kinect should accept phrases at a faster rate.
    Tuesday, April 14, 2015 8:35 AM