Waseda University’s Kobayashi Lab has published a video of their conversational robot ROBISUKE giving some directions. Two things are rather striking about this presentation when compared with other speech recognition / speech synthesis demonstrations. First, you’ll notice how quickly ROBISUKE responds. This has to do with the lab’s emphasis on conversational flow; ROBISUKE doesn’t awkwardly pause or talk over its conversational partner.
The second surprise is ROBISUKE’s response to intonation (such as the varying uses of “hai“) and context; when the name of the street is repeated by its partner, ROBISUKE adds detail to clarify which street it is referring to. It’s difficult to say how much of this was scripted, but if taken at face value it is quite impressive indeed. This also happens to be the first video footage of this robot I have seen online.
[source: Waseda University Kobayashi Lab (JP)]