Basically, you'd need some sort of biometric API \ Voice analysis
You'd need to record the wav file, then store that in the database against a user account.
Then, when user wants to log on, record the new audio from a microphone at the terminal, send that to some sort of service that compares the new recording agaist the database version and performs spectral analysis matching of the 2 recordings.
This really isn't a small requirment!
However, some tools that may help you
http://www.exocortex.org/dsp/[
^]
Speech recognition using exocortex
http://www.generation5.org/content/2004/noReco.asp[
^]
Update:
What you are talking about is a biometric logon by using voice print audio.
E.g. Say the logon was 'password' for User 1, if User 1 said 'password' and User 2 said 'password', their voice prints would be different enough so that only User 1 would actually get logged on
The second link I posted is doing a lot of this. You'd need to create a
spectrogram [
^] to perform the analysis.
There's no quick or easy answer to this. You've got some serious coding to do!