Another German speaker here. I agree with the general impression in this thread: It's a cool concept, but I can't see anything revolutionary in the execution. Basically, it looks like they plugged a speech recognition engine into a translation service which in turn sends its output to a text-to-speech engine. None of these components perform horribly bad, but neither are they any better than the competitors' offerings.
Natural speech processing seems to work comparatively well for English but not so well for German, hence the German lady's strange speech pattern. (Or they just wanted to make extra sure that there were no recognition errors showing up in the English translation.) The English->German translation itself seems to be as accurate or inaccurate as Bing Translate. Some sentences are almost spot on, others are a complete mess. Generally, it's the same quality you get from any current translation engine. You just can't tell from the German->English translation, since they used extremely unnatural German phrasing to give good English results.
All in all: Nice idea, but Google could probably put together something almost identical in a very short time. They already have all the parts they need, they just need to pipe output from one to the other. Apple isn't quite there, though AFAIK they don't have a suitable translation engine. (Next on Tim Cook's shopping list, maybe?)
