This paper introduces a novel unsupervised approach to semantic role induction that uses a generative Bayesian model. To the best of our knowledge, it is the first model that jointly clusters syntactic verbs arguments into semantic roles, and also creates verbs classes according to the syntactic frames accepted by the verbs. The model is evaluated on French and English, outperforming, in both cases, a strong baseline. On English, it achieves results comparable to state-of-the-art unsupervised approaches to semantic role induction.