Author |
Topic: How it works? (Read 1812 times) |
|
Slava
YaBB Newbie
I love YaBB 1G - SP1!
Posts: 2
|
 |
How it works?
« on: Mar 10th, 2009, 8:57am » |
 Quote  Modify
|
Hi, Denis. I have tried you voice recognition screensaver and I have admired by it. I have not seen any program which does something like this. Can you provide some information about principes your program using? I have read this article http://www.okob.net/projects/vscrsvr/design.htm but it stills some questions for me. Like "how spectrograms are buid?", "what is an input for the neural network?" etc. Can you provide more detaied information (like links to external resources, links to books or scientific researches you have used for this project).
Thank You!
|
|
 IP Logged |
|
|
|
Denis Bakin
YaBB Administrator
Gender: 
Posts: 163
|
|
Re: How it works?
« Reply #1 on: Mar 10th, 2009, 4:27pm » |
Quote Modify
|
Hi Slava,
It's been quite a while since I wrote that app. I'll check if there still are notes w/ references left, but it is not very likely I'd find many and anyway that would be outdated.
The voice print is built by splitting the recorded sound sample into multiple short overlapping segments and running FFT (Fast Fourier Transform) on each of them.
You can find details in the article about "Voice Print/Spectrograms" on Wikipedia: http://en.wikipedia.org/wiki/Spectrogram
The voice print data for the sound sample are scaled and fed to the Back Propagation NN for recognition.
You should be able to find plenty of information for "Back Propagation" on Internet too. For examplle, check this Wikipedia arcticle:
http://en.wikipedia.org/wiki/Multilayer_perceptron
That is what it does in the nutshell, but you have probably already figured that from the online paper.
If you let me know what particularly you are interested in I'll try to give you more details.
Hope that helps,
Denis
|
|
IP Logged |
---------------------------
Denis Bakin
|
|
|
Slava
YaBB Newbie
I love YaBB 1G - SP1!
Posts: 2
|
|
Re: How it works?
« Reply #2 on: Mar 11th, 2009, 6:27am » |
Quote Modify
|
Hi, Denis,
Thanks a lot for provided information, this is exactly I am looking for.
Good luck!
Slava
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print