Megaman,
This is a fun an interesting project.
I tried something like this a long time ago, on a commodore 64. unfortunately the 2mhz processor didn't allow for a quick enough sample rate, and I wasn't able to store a lot of voice finger-prints. I wound up rigging a Microphone (to the Joystick input port).
Now I believe that this would still work today with better results. I'm not much into sound anymore so I will have to check the docs to see if DB allows acces to the mic input. If not you could always wire it up to the joystick port.
You can then graph the voice wave. Compare the graph to base patterns to see if there is a close match. The difficult part is establishing good base patterns, adapting those patterns to a specific user(persons) voice profile so that the program can learn to identify the persons voice. At the time I attempted this I went for ful words rather than phonetic learning, which is obviously a better method.
I also ran into issues with timing and reshaping the input waves(for example if someone spoke faster or slower than my base pattern).
Even though my experiments were hardly accurate enough for anything useful, it was interesting. I moved on to try programming a lie detector (again completely unsuccessful) that would measure moisture on the palms of hands (through joy port 2), and voice abnormalities. I think a magic 8-ball was more reliable.
Good luck with this. I think it can be done. In fact, building the interface to the joyport was half the fun. So you may decide to try it.
