Okay, mesh work starting to get a bit out of my realm.
Making Vocaloids try to do everyday speech in Japanese is a really hard process. English is.... slow.
I've been working on getting classic Miku just to do the single-word callouts (translation, contact, etc.) but it's a slow process - since human speech is so intricate in terms of pitch and dynamics, ultimate you just end up saying the line out loud over and over until you get a feel for the pitch and volume changes, then try and mimic them with the Vocaloid's parametres.
Not a fast process, but slowly coming along.
------------------
EDIT: Okay, I finally got one phrase tuned to the best possible quality I can. The people I talk in the Vocaloid communities on both sides of the Pacific were right: Miku is VERY hard to do English speech with, and is pretty much impossible to get a really high-quality speaking voice out of without a dedicated English voicebank. Mind you, one is on the way, but until then, Luka is an option because she has an English voicebank already, and maybe Gumi, since she has a very natural tone as it is and is very easy to work with.
But you're both trying to get a vocal sequencer to do normal speech, AND you're trying to morph Japanese syllabary into English, which isn't easy at all, given the limitations of the Vocaloid software - remember that it's a sequencer, not a synthesizer.
Either way, I tested it with the hover doors closed warning, and here are my results, the best I could possibly make them.
http://www.youtube.com/watch?v=whe6UyEMC04