Lalilo: a reading assistant for children featuring speech recognition-based reading mistake detection

Corentin Hembise (1) , Lucile Gelin (2, 1) , Morgane Daniel (1)

1 Lalilo, Paris
2 IRIT-SAMoVA - Équipe Structuration, Analyse et MOdélisation de documents Vidéo et Audio

Lalilo is a reading assistant intended to help kindergarten to second grade students to master their reading skills. Students progress at their own pace thanks to an adaptive learning system that differentiates instructions. Teachers can access data on their students' progression. Among other exercises, a readaloud exercise is provided for students to practice their reading. This exercise uses a reading mistake detection system based on speech recognition to offer automatic feedback on the child's reading. Since speech recognition on children learning to read is highly challenging, we overcome potential inaccurate thus damageable feedback with an uncertainty estimation leading to a neutral feedback.

Mots clés: reading mistakes detection; child speech; reading assistant; uncertainty estimation

Domaines: Traitement du signal et de l'image [eess.SP]; Multimédia [cs.MM]; Intelligence artificielle [cs.AI]

Corentin Hembise, Lucile Gelin, Morgane Daniel. Lalilo: a reading assistant for children featuring speech recognition-based reading mistake detection. Annual Conference of the International Speech Communication Association (INTERSPEECH), Show & Tell contribution, Aug 2021, Brno, Czech Republic. ⟨hal-03410194⟩