The rather remarkable video above is a demo of an iPhone speech translation app from a company called Sakhr. You say a sentence (in either language), and the app converts your speech to the language's text, translates into the other written language, then synthesizes the second language speech.
(It's interesting that, in this demo, the synthesized Arabic speech seems to have more "realistic" inflection than its English counterpart…)
As someone who's day job involves a LOT of narration recording — and monthly visits to my boss, whining "What's the latest in text-to-speech software?" — I can safely say that the available technology is far from cutting edge. (Adobe Captivate, for instance, offers T2S as a half-gigabyte bonus download of code and audio snippets, and the results STILL sound pretty robotic.) Sakhr — who specializes in Arabic T2S, and counts the US Departments of Defence and Homeland Security as customers — at least seem to be playing to their core strengths.
The app, by the way, is not commercially available.