A Language Acquisition Support System that Presents Differences and Distances from Model Speech

It is difficult for language learners to know whether they are speaking well and, if not, to know where their speech differs from that of native speakers and how much their speech differs from that of native speakers. Therefore, we propose a novel language learning system that solves these problems. The system uses self-supervised learning to determine whether the user’s speech is good or not. The system also shows where the user’s speech differs from the native speaker’s speech by highlighting the corresponding places on the user’s speech waveform. It also represents the learner’s speech and the native speaker’s speech as points on a two-dimensional coordinate to show the user how far apart they are. We expect that the learner’s speech will gradually become better by repeatedly modifying the speech to eliminate these differences and bring the distance closer.