After many matlab simulations and hours of research (Adam 
found some great speech processing books that are helping us 
mucho) we have redirected our efforts.  Initially, we thought that
wavelets would be the panacea that would allow us to easily 
compare speech of different words and speakers.  However, it 
appears that sampling at 8K (or even 16K) and applying filterbanks
to the sample did not give enough frequency resolution to 
differentiate between speakers.  We will still try and use a 
filterbank approach in analyzing words with transient elements 
(sh and ch sounds) but for vowel sound we have began focusing on 
pitch determination and formant analysis.
   As we have learned from the many speech analysis books we have
aquired, the human vocal system can be modeled as an all-pole
system when producing vowel sounds.  Using forward-backward
Auto-Regressive methods we were able to identify the formant
components in vowel sound sample.  We hope to be able to use the
first two formant frequencies and amplitudes in determing the
vowel spoken.  As shown in many text, the first few formant
frequencies when plotted against each other tend to cluster. 
This would be useful in producing formant-vowel maps and
comparing test signal formant against the map in determing the
sampled vowel.  
   Another useful idea in speech recognition is the vocal tract
model.  We were able to produce an AR model (using the ar command
in matlab) and then use the AR transfer function excited by an 
impulse train (representing the glottal impulses) to SYNTHESIZE 
the vowel sound!  We thought that was super kewl.  
   The gang is slowly but surely regrouping after a difficult
start and will pull together a report outlining the difficulties 
we've encountered and the progress we've made in speech synthesis.
The formant analysis we are currently learning about looks 
promising.