Helsinki-2025

At NTNU, we have developed an online platform for pronunciation training called the Computer-Assisted Listening and Speaking Tutor (CALST). This is a multilingual platform both in the sense that any language can be implemented as a target language and in the sense that it tailors exercises to the learners’ native language.

So far, we have implemented Norwegian, English, Greek, Italian, Spanish, and Catalan as target languages (L2). For each language, a comprehensive set of exercises has been developed, including exercises for individual sounds, consonant clusters, stress, and (in Norwegian) lexical tones. But to use their time efficiently, learners should focus on exercises for the mismatches between L1 and L2. We use L1-L2map to find these and tailor the exercise selection to the learner.

CALST offers listening (discrimination and identification), speaking, and spelling exercises. In the present version of CALST, learners evaluate their own pronunciation. For many learners, this raises the awareness about the acoustic differences between their own pronunciation and that of the tutor. But not all learners are so perceptually acute, and these would benefit from an automatic pronunciation assessment.

We have previously informally tested automatic pronunciation assessment using simple HMM and DTW approaches, but results at that time were not considered sufficiently reliable for implementation in CALST. We would welcome a discussion about the possibilities of implementing more state-of-the-art, scalable automatic pronunciation assessment which can give targeted phonetic feedback to the learner about the sometimes subtle differences that are distinctive in the target language.