Original Proposal (boy were we naive)

First there were punch cards. Then keyboards. Next came the mouse. The exponentially growing power of computers may call for a new, more elegant way of inputing data. Wouldn't it be nice to say "Computer, Open Netscape", or dictate a paper into a text editor instead of typing?

We plan to use wavelets in exploring speech interpretation. The continious wavelet transform can be used to transform a time varying signal (speech audio in our case) to a time frequency domain. One of the advantages we gain by using the CWT over traditional Fourier Transform is that time information is more readily avaliable. We hope to use this time information to process different voices saying strings of words.

Using Matlab extensively, we will build (program) an interpreter. We will transform speech signals of words like "cat", "hat", "rat" into time-frequency representations and see how well we can differentiate the words. We will also explore our interpreters ability to differentiate between speakers. Finally we hope to develop a process that will be able to tell when a few words are spoken, what those words are, what order were they spoken in, and who spoke them.

Probelms we hope to tackle:

1. Recognizing words
2. Recognizing a string of words
3. Identifying a voice