Verbmobil is a speaker-independent system that offers translation assistance in dialogue situations. In co-operation with other institutes we are developing the speech synthesis module within Verbmobil for German and American English. Current priority is given to an enhancement of naturalness of our PSOLA based concatenative synthesis of German. Due to a tight schedule we investigated alternative methods to our traditional

approach. In our opinion, quality enhancement of PSOLA based concatenative synthesis has reached its limits.

We decided to avoid concatenation points and prosodic manipulations as much as possible. Our new approach obtains

prosodic diversity by using those synthesis units which inherently possess the necessary prosodic features. To get fast results we started with words as primary synthesis units. The outcome is encouraging. Even a first version of our system frequently succeeds in synthesising utterances with close to natural quality.