5 replies to this topic

[achinapong]

I got stuck to find the Neural Networks Paradigm.

I'm seriously want to use Spatio-Temporal Pattern Recognition(Avalanche) in my project "Recognize Emotion in Speech"
but there's no paradigm of that networks I have ever seen.

Due to I use Joone as an engine. I must wire each perceptron my self.

Anyone have any Idea?


Warmest Regards to Everyone
Chinapong

[NomadRock]

Ok, it is fairly obvious to me that you do not understand the terms you are using, and are simply copying and pasting interesting things.

First off, what are you doing this project for, and why do you think you need to use this type of network. Perhaps you can do a simple problem that you already understand?

Is using Joone a requirement of your class, or is this just the first one you came up with on a google search?

I am happy to help, but I think you are in over your head, and you need to step back and slow down a bit.

[achinapong]

First of all, Thank you very much for your advice. I'm in hurry due to the time limit(within January). Your advice really help me back to my sense thank you.



About Project. It's about to recognize user emotion by using speech as input.

- I've specify input as a 16-bit PCM 22050 Hz Wav file.
- Emotion I need to recognize is neutral happy angry and sad.
- I'll use the pitch information from file (mean , s.d. ,range,frequency)as a training set and testing data.
- When the networks are finish. I'll use it as user interface engine on various platform (Linux,WindowsXP,and Mobile). THat's why I thing that my best choice is Joone that use java as core engine


May be I'm wrong but I think that recognizing speech from computer user is
time series pattern recognition work. That's why I think that Spatio-Temporal paradigm is best match for me.

I'm still only a newbie on AI. I've learn only the basic of neural networks and some simple paradigm like feed forward and basic of multilayer perceptron.



Could you please give me some advice about my project? I'm almost gathered all raw data I need to use. I'll be ready to start training AI within a few weeks.


Thank you for everything,

Warmest regards,
Chinapong

[NomadRock]

Is the sound file always the same length? A standard multilayer perceptron could quite possibly do the trick for you.

Most of the advancements have come in helping us to get neural nets to converge faster, in otherwords getting them fully trained using less trials. If you are not worried about speed so much, then many of these more complicated nets may not be worth your time to learn them while you are on a time budget to get this project completed.

[achinapong]

Sorry for slow reply, I've got lots of work to do in my school Sportday.


My sound files are in same length (5 seconds). Speed isn't the important thing. I'm also interested in the advancement that help networks fully trained by using less trials too. Could you give me more info about these advancement? I will try using Perceptron networks.



Thank you for your help,
Warmest Regards,
Chinapong

[NomadRock]

Well many of them get very complicated, and are very difficult to understand. I suggest you get a multilayer perceptron working before you go more complex. Much like you should learn to multiply and divide before trying to learn integrals.