I've noticed the chronic issue of someone acting 1500 voiced lines. Apart from driving the actor insane (take 12 of line 1422 ) and hoarse in voice, it's a bit unrealistic. I concur that the major lines can be voice acted, more accurately put I suggest scenes that require inflection not just casual conversational tone. I would suggest a slightly different approach to the esoteric scenes that aren't required to do anything with much emotion. Voice synthesis using the actors voice and some slight tuning something like espeak
could be used to make esoteric speech (IE stuff that doesn't involve inflection etc.)
I have tested this with the generic MBROLA diphones provided with the program and the results were reasonable, for both male and female gender. I would suggest this as a reasonable means to an end. Especially for the casual conversation in the game. With some diphonic tweaks you can create a considerably larger set of voiced lines, and reduce the voice actors 'wear'. This is what you have to do in any case (without the addition of speech synthesis), so I see it as a creative means to getting there faster.
There is an incomplete SSML interface on the synthesizer but to key you in IT WORKS, I've USED IT, for a simple project of giving personcom CM partners in Oblivion a somewhat artificial voice instead of a copy of Imperial race voices. Worked surprisingly well that I didn't erase all evidence I tried it. I still use it too. LOL
I wouldn't recommend it if I hadn't tried it already. The Carnegie Mellon speech website is fairly good, that is used by Microsoft (who shamelessly gives no credit to the creators as usual), they just used professional voice actors and tuned things a bit to make the Microsoft speech tools. Also there voice recognition software is a copy of the CMU sphinx tool kit. (blatant again but the licensing allows this). Just tweaked specifically for Microsoft products. The issue is becoming familiar with the tool. I suggest espeak because it is much simpler to use and fairly straight forward. No need to compile anything for using the synthesis tool.
This would allow a lot of 'blah blah' lines to be generated without having someone speak them, and leave the more important lines to be tweaked and worked on by the voice actor. In addition any 'normal' lines can be generic or voice acted and or tweaked as needed.
Just a thought.