This paper describes an approach which allow a humanoid robot to automatically acquire vocalization capability by learning from a human tutor. The proposed algorithm can, at the same time, synthesize speech utterances from unrestricted text and generate facial movements of the humanoid head synchronized with the generated speech. The algorithm uses fuzzy articulatory rules, derived from the International Phonetic Alphabet (IPA) to allow simpler adaptation to different languages, and genetic optimization of the membership degrees. Experimental results show a good subjective acceptance of the acquired vocalization in terms of quality, naturalness and synchronization. Although the algorithm has been implemented on a virtual talking face, it could eventually be used also in mechanical vocalization systems.
