Most second language (L2) speakers speak a foreign language with perceivable traces of a foreign accent. Nonnative variants in intonation, pitch range, fluency, speech rate, pauses and rhythm are reliable predictors of foreign accent. Little attention is paid to teaching L2 prosody in classroom context. L2 learners, instead, may greatly benefit from computer-assisted pronunciation training (CAPT) systems to improve the perception and production of L2 suprasegmentals. For this project, an existing database of prosodically manipulated utterances in L2 is available and analysis can be carried out to examine in which acosutic features L2 leaners improve aafter the training and to what extent the acoustic convergence towards the native speakers is perceptually salient to native listeners. Alternatively, a new corpus of L2 speech can be collected and various natuve models can be used for pronunciation training to identify the better model-learner match.