Towards an enhanced prosodic model adapted to dialogue applications