Natural appearance and behavior is most important for humanoid robots that are predestined for human-robot interaction. One aspect of natural appearance is the generation of realistic mouth motions. In this thesis a system for generating realistic mouth motions for humanoid robots was developed. Requirements for this system are: no limitation to a specific speech source or speech synthesizer and adaptability to different robot platforms. Therefore a combined approach using an audio data method and a phoneme method had been implemented and a generic interface for the different speech sources had been developed. The system has been integrated and tested on the humanoid robot ROMAN.