We have designed automatic feedback for the realisation of the prosody of a foreign language. Besides classical F0 displays, two kinds of feedback are provided to learners, each of them based upon a comparison between a reference and the learner's production. The first feedback, a diagnosis, provided both in the form of a short text and visual displays such as arrows, comes from an acoustic evaluation of the learner's realisation; it deals with two prosodic cues: the melodic curve, and phoneme duration. The second feedback is perceptual and consists in a replacement of the learner's prosodic cues (duration and F0) by those of the reference. A pilot experiment has been undertaken to test the immediate impact of the "advanced" feedback proposed here. We have chosen to test the production of English lexical accent in isolated words by French speakers. It shows that feedback based upon diagnosis and speech modification enables French learners with a low production level to improve their realisations of English lexical accents more than (simple) auditory feedback. On the contrary, for advanced learners involved in this study, auditory feedback appears to be as efficient as more elaborated feedback.