Her voice is highly processed using formant shifting and manipulation of artifacting to get it to kind of slide up and down but still be robotic, demonstrated in the excellent video linked by Tom.
If you know any music production techniques or programs, you could get a similar effect by using the Mac's built in speech synthesizer and an Auto-tuner similar to what you hear on pop radio nowadays. Instead of automatically figuring out what pitch you're trying to achieve, you'd need to "force" it from one pitch to another by feeding the Auto Tune program a parameter - so if you're talking in the pitch of C, you get the GLaDOS effect by forcing it up to G or A.
The types of plugins used for this sort of processing are developed by Antares and Melodyne; audio professionals use these the most. You could even enhance the vocal effect even more by using an audio device called a compressor, which in context would make the volume of her voice constant like the pitch.
In the video that was posted using content from Portal 1 as an example, the pitch is determined by mouse click, and it is being operated on a file that has already been recorded. I know for a fact that you could achieve this same effect in real-time with a MIDI keyboard, the aforementioned Auto Tune plugin, and a host such as Ableton Live.
Using the keyboard, you can select a note that you would like to tune to. Then, you can sing or speak well below that tone and force the plugin to correct your voice. It uses a synthesizer to fill in the tone that you don't have, so the farther you are from the pitch, the more synthesizer there is, and the more robotic it sounds. When it messes up, you get "artifacting" which is a distortion caused by the plugin - you can hear that in GLaDOS' voice.
So as you're speaking, if you hit different notes on the keyboard, your voice will change pitch, giving it a little dynamics. Hope that makes sense, and its definitely a fun project to go into if you've never worked with audio before.