Multi-talker dialogue synthesis