Multidimensional Assessment of English Speech on Pronunciation, Fluency and Prosody
Zhigang Ou, Yun Gao, Jianfeng Cheng · 2021
The assessment of English speech is critical for learning and teaching. Existing assessment methods focus on scoring the phonation accuracy as the quality of articulation, which merely represent single evaluations of correction on speech. In this paper, we propose a novel multidimensional assessment system that takes into account linguistics and perception characteristics of speech. The system helps English learners rectify their pronunciations through estimating phonetic and prosodic patterns. We produce a dataset with dense annotations and develop a series of modeling methods to compute word-level pronunciation scores. Further, this work investigates whether these scores are salient indicators of quality on fluency and prosody. Meanwhile, this study innovatively notes and explores new contours and metrics to implement evaluation process. We conduct qualitative and quantitative experiments, demonstrating that our system presents competitive and instrumental estimation techniques on English speech.