none
Can I reduce the noise of the television sound [ wind or etc. ] at kinect recognizability. RRS feed

  • Question

  • I can use speech engine.

    So I can say a word  "kinect" that use motion recognition sensor.

    But kinect by television that makes some sound blah-blah isn't recognizing.

    How I can reduce the noise?

    Or How I can recognize my word to avoid that noise.


    Thursday, November 8, 2012 10:17 AM

Answers

  • The Kinect needs to be able to hear you over the television.  To be more specific, the Speech recognition engine needs to hear you about 20dB louder than it hears anything else.  Sound energy drops off with the square of the distance from the sound source.  (Twice as far away translates to 1/4 the sound energy), so the Kinect next to the television is likely to hear the TV *significantly* louder than it hears you if you are farther away from it.  Given that the computer running the Kinect doesn't know what sound is playing, it is *very* hard (read close to impossible) to separate out the sound that's coming from the television, or to recognize the sound from the person. 

    That having been said, there *are* some things that you can try (aside from turning down your TV):

    • Do manual beam steering, and point the beam at the tracked skeleton of the user who is speaking.  The first thing to get confused in a loud environment is the audio module responsible for deciding which direction to listen in.  It hears the TV, the TV is loud, so it thinks it should try to listen to the TV.
    • Position the Kinect above the TV and speakers, and in front of the speakers (as in closer to the user than the speakers are)  The individual microphones in the Kinect have some intrinsic directionality, and the housing that holds them provides some acoustic damping.

    Aside from that, the default settings on the Audio source should be doing the best that can be done. You could also build a custom housing for the Kinect that provided better acoustic isolation, but that would definitely look a little funny sitting on top of your TV. :)

    Saturday, December 1, 2012 8:26 AM

All replies

  • The Kinect needs to be able to hear you over the television.  To be more specific, the Speech recognition engine needs to hear you about 20dB louder than it hears anything else.  Sound energy drops off with the square of the distance from the sound source.  (Twice as far away translates to 1/4 the sound energy), so the Kinect next to the television is likely to hear the TV *significantly* louder than it hears you if you are farther away from it.  Given that the computer running the Kinect doesn't know what sound is playing, it is *very* hard (read close to impossible) to separate out the sound that's coming from the television, or to recognize the sound from the person. 

    That having been said, there *are* some things that you can try (aside from turning down your TV):

    • Do manual beam steering, and point the beam at the tracked skeleton of the user who is speaking.  The first thing to get confused in a loud environment is the audio module responsible for deciding which direction to listen in.  It hears the TV, the TV is loud, so it thinks it should try to listen to the TV.
    • Position the Kinect above the TV and speakers, and in front of the speakers (as in closer to the user than the speakers are)  The individual microphones in the Kinect have some intrinsic directionality, and the housing that holds them provides some acoustic damping.

    Aside from that, the default settings on the Audio source should be doing the best that can be done. You could also build a custom housing for the Kinect that provided better acoustic isolation, but that would definitely look a little funny sitting on top of your TV. :)

    Saturday, December 1, 2012 8:26 AM
  • Is it possibile to use the EchoCancellationSpeakerIndex parameter ?

    And redirect TV input to Kinect's computer then transform the input into an output for EchoCancellation ?


    Saturday, August 17, 2013 11:41 PM
  • Please post a new question and link to previous discussions if they are relevant. If you are familiar with the Windows Audio API's for AEC and voice cancelation, it is possible to some degree. It depends on what exactly your scenario is and as Chris stated whether the volumes can be overcome.

    See the samples provided in the toolkit for using Kinect with Windows Audio.

    http://msdn.microsoft.com/en-us/library/jj883683.aspx

    http://msdn.microsoft.com/en-us/library/jj883687.aspx


    Carmine Sirignano - MSFT

    Monday, August 19, 2013 8:27 PM