QUOTE(jcss @ 6 Mar, 2009 - 12:01 PM)

I'm doing a speaker verification system. I don't want any solution to building such a system, but like to know how other people have implemented it, since there aren't a lot prevailing implementation code about this.
By "speaker verification", I assume that you wish the system to verify that the speaker is indeed a particular person on a list of identified users. If so, then I would sketch out the following general plan:
1. Establish a standard phrase which everyone will utter for verification, like "My voice is my passport, verify me."
2. Make multiple recordings of both identified users and other people.
3. Based on extracted features of these recordings (spectral transforms, statistical summaries, etc.), either:
3A. Establish some means of index people's voices (perhaps as a hash function)
or...
3B. Construct a classifier (linear discriminant, neural network, etc.) for each user vs. "everyone else".
4. Test carefully, noting that a reasonable sample of people not known to the system should be tried (and they should obviously be rejected).
Good luck!