Automatic pronunciation assessment has the potential to play a key role in the process of improving proficiency in a foreign language. The process represents a modern, smart learning alternative where students can receive immediate feedback with very low friction. As an overview, such systems are built on top of regular Automatic Speech Recognition engines using data collected from various native speakers but operate at a lower level, where the system recognizes phonemes instead of individual words. Another key difference is that such systems do not attempt to perform automatic corrections using a language model; instead, the output is a measure of resemblance with the learned baseline with emphasis on the detection of mispronunciations. In this study, we introduce pROnouce, a tool designed for the Romanian language, which also considers gamification to ensure a more pleasant experience. Two approaches for pronunciation assessment were considered, both using Deep Neural Network models, coupled with a method to further expand the training dataset using recordings for other languages that share the phoneme set. Our system was evaluated by more than 150 individuals at Expo Dubai 2020 who were interested in experimenting with introductory Romanian words and provided feedback.
CITATION STYLE
Ungureanu, D., Ruseti, S., Toma, I., & Dascalu, M. (2023). pROnounce: Automatic Pronunciation Assessment for Romanian. In Smart Innovation, Systems and Technologies (Vol. 908, pp. 103–114). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-19-5240-1_7
Mendeley helps you to discover research relevant for your work.