In recent years, Computer Assisted Pronunciation Technology (CAPT) systems have been developed that can help Japanese learners to study foreign languages. We have been developing a pronunciation training system to evaluate and correct learner's pronunciation by extracting articulatory-features (AFs). In this paper, we propose a novel pronunciation training system that can plot the place and manner of articulation of learner's pronunciation on an International Phonetic Alphabet (IPA) chart in real time. First, the proposed system converts input speech into AF-sequences by using multi-layer neural networks (MLNs). Then, the AF-sequences are converted into x-y coordinates and plotted on an IPA chart to show his/her articulation in real time. Lastly, we investigate plotting accuracies on the IPA chart through experimental evaluation.

Index Terms: pronunciation training, articulatory feature, IPA chart