Analyze the Influence of Features Extracted from Audios on the Results of IELTS Speaking Test

The prominence of international English certifications such as IELTS, TOEIC, and TOEFL has surged in Vietnam's education and job sectors. Despite common self-practice in reading and listening, speaking and writing skills necessitating supervision, access to reputable preparation centers is restricted, particularly in rural regions where long-term tuition fees pose barriers. Moreover, instructors encounter difficulties in correcting speaking errors for multiple students, underscoring the demand for an automated assessment system. Our study addresses a gap in research by aiming to facilitate English language learning for Vietnamese. By utilizing data from a test preparation center in Ho Chi Minh City, we extract meaningful features from audio recordings of speaking tests. Subsequently, we do statistical analysis to elucidate the influence of these features on IELTS speaking scores. Furthermore, Machine Learning algorithms are employed to automatically predict scores based on features extracted, showing promising results in predicting speaking scores. Our analytical researches help Vietnamese learners obtain useful insights about speaking criteria assessment and their weaknesses in the IELTS-speaking test. Therefore, Vietnamese English learners and teachers can choose the appropriate strategies to improve their scores.