none
Rejected "RetainedAudio_*" RRS feed

  • Question

  • Just out of curiosity, what is the general idea for having the DumpRecordedAudio method in the Speech(cs) sample?  I understand what it does but in what situations would you and/or your team members think of using it?  In the speech recognition sense, I think it's a great example of getting data back to the person writing for a user.  Just interested.

    Also, thanks to anyone involved for the work on the kinect sdk.  This is a lot of fun and I plan on using what I learn here for class projects.  Again, thanks.

    Saturday, June 18, 2011 6:30 AM

Answers

  • Thank you for the encouragement!

    We thought it would be helpful for debugging/exploration purposes. The three recognition choices in sample are "red", "green" and "blue", and you might say "blue" and have speech engine reject this speech event. So, you're confused as to why it was rejected. By having the captured speech sample corresponding to rejection you can listen to it and try to determine whether the recognition error was caused by a low quality sound capture, e.g.: maybe there was a loud noise in the background at the same time as you said "blue" which you didn't notice before, or for whatever reason the sound sample doesn't actually sound like the word "blue".

    Having the rejected samples can hopefully help you understand better how speech is being recognized and where common problems will arise.

    It's also a hint for developers that RecognitionResult type includes an Audio Property that can be used to write a WAV file with few lines of code.

    Hope this helps,
    Eddy


    I'm here to help
    Monday, June 20, 2011 9:42 PM

All replies

  • Thank you for the encouragement!

    We thought it would be helpful for debugging/exploration purposes. The three recognition choices in sample are "red", "green" and "blue", and you might say "blue" and have speech engine reject this speech event. So, you're confused as to why it was rejected. By having the captured speech sample corresponding to rejection you can listen to it and try to determine whether the recognition error was caused by a low quality sound capture, e.g.: maybe there was a loud noise in the background at the same time as you said "blue" which you didn't notice before, or for whatever reason the sound sample doesn't actually sound like the word "blue".

    Having the rejected samples can hopefully help you understand better how speech is being recognized and where common problems will arise.

    It's also a hint for developers that RecognitionResult type includes an Audio Property that can be used to write a WAV file with few lines of code.

    Hope this helps,
    Eddy


    I'm here to help
    Monday, June 20, 2011 9:42 PM
  • Awesome, thanks!
    Friday, June 24, 2011 2:30 AM