Thoughts on Speech Recognition
Principal Component Analysis (PCA) is a method of capturing the variance in a signal. Using PCA in the context of speech recognition is good for separating multiple components from a blind source, or one unknown signal. The reason that speech recognition is difficult is that it requires a controlled environment to work. Any change in the rate of the recorded sound, someone who talks fast compared to someone who talks very slow would change the principal components' variance and the sound would not be recognized. Wavelet analysis might lead to a solution that compensates for changes in frequency in addition to multiple time resolutions.
hummm...
Sounds interesting. You could add a small readme to the program that told htem how to talk etc. You could use it to instruct a creature to do things in a simulation, really interesting you should try to make one! :D
Re: FWT
Actually worked out a wavelet transform a lot of possibilities!
I have not failed! I have only tried 100,000 ways that will not work. —Thomas Edison
Oh really!? Could you post
Oh really!? Could you post it on a website so I could see?...or is it still in the making???
~signture~
"The planet is dieing and we shall see the final moments of this beautiful world fade out of existence while we sit on throne's of twisted metal." --Quote'd from The Writing Game, Jui.
Negative on Statistical Analysis
I think the primary reason industry goes for PCA is that it works well with current DSP applications. Wavelets with multiple delays (e.g. Maybe a Mean Field Approximation) that could compensate for multiple shifts in frequency. All theoretical.