Voice Recognition

  • Thread starter Thread starter Caroline
  • Start date Start date
C

Caroline

I need to recognize spoken digits and a few letters in my pocket pc app.
Anybody out there
ever done any speech recognition in compact framework?

Thank you in advance for any information,
Caroline
 
Just need to expand on the requirements:
- This is not a web services app, so I need recognition running on the PPC.
- It needs to understand different kinds of voices.

I just started doing some research on this and so far I can't find any
sample code for that.

Your advice would be appreciated.
Thank you,
Caroline
 
It's a tall order and I've not seen anyone implement it and share the code.
Conceptually I know how you'd attack it, so I can at least point you where
I'd go if I had to do it.

1. Use the OpenNETCF.Multimedia.Audio.Recorder to collect the sound. This
will give you easy access to the audio data itself.
2. Run the audio data through one of the OpenNETCF.MathEx.FFT classes to get
a spectrum on it. You might have to do an FFT over time to get a
"fingerprint" on the sound (i.e. don't FFT the entire sound, but maybe a
running FFT of a 128 or 256 data point window).
3. Once you have this you can generate a spectrum for a given word. Have
several people say the word several times. Try to pick a wide array of
speakers.
4. Now the fun - you need to grab a spectrum from a speaker and try to
match it to a known spectrum. If you have several spectra for a given word
you might try a closest match or a bounds fit. Who knows, maybe simple peak
analysis would be enough - looking at the data is the only way to know.

-Chris
 
Thank you Chris. That sounds like a good idea, a little more work but
interesting and challenging. I have never done anything with spectrums, so I
have to do some research on that as well. I've seem Casey Chestnut's "from
scratch" article, really admire his work, too bad he won't share his code,
but I understand his point of view. The phonix solution seems like it would
work, but they won't even give out the pricing on their website so I imagine
it could be quite expensive.

So far I think your idea is the best. Thanks for sharing, and I will post if
I find anything interesting.
 
Thanks. I have looked at the article, good ideas. I also found a book that
looks promising:
Spoken Language Processing: A Guide to Theory, Algorithm and System
Development (Paperback)
by Xuedong Huang, Alex Acero, Hsiao-Wuen Hon


Sergey Bogdanov said:
... in addition, Casey Chesnut described here some ideas:
http://www.mperfect.net/noReco/


--
Sergey Bogdanov [.NET CF MVP, MCSD]
http://www.sergeybogdanov.com

Just need to expand on the requirements:
- This is not a web services app, so I need recognition running on the
PPC.
- It needs to understand different kinds of voices.

I just started doing some research on this and so far I can't find any
sample code for that.

Your advice would be appreciated.
Thank you,
Caroline
Click to expand...
Click to expand...
 
Back
Top