MEASURING THE IMPACT OF VIDEO DUBBING ON EFL LEARNERS’ INTELLIGIBILITY: A WER-BASED ANALYSIS OF NON-NATIVE ENGLISH PRONUNCIATION IN ZOOM’S AI-POWERED ASR

Abstract

This study investigates the impact of Video Dubbing (VD) on the intelligibility of English pronunciation among Indonesian EFL learners from Sundanese and Javanese backgrounds. Using a convergent parallel mixed-methods design, the research analyzes non-native English pronunciation via Word Error Rate (WER) metrics from Zoom's AI-powered Automatic Speech Recognition (ASR) system. Data from Chapter IV of the dissertation covers pre- and post-VD assessments, following an Analyzing–Tailoring–Modeling (ATM) protocol that integrates audiovisual modeling, ASR-generated captions, and acoustic feedback. Quantitative analysis showed a significant improvement in intelligibility. Pre-intervention WER averaged 28.5%, reflecting L1 interference. Post-VD implementation, WER decreased by 42% to 16.6%, a large effect (Cohen's d = 1.12), and listening comprehension scores rose from 62% to 85% accuracy (η² = 0.34, p < 0.001). Acoustic analysis, performed using Praat, revealed enhanced vowel reduction (schwa durations shortened by 35%, from 120 ms to 78 ms) and improved prosodic alignment (mean F0 variation reduced by 18 Hz). Praat parameters included LPC (Linear Predictive Coding) analysis (order 10–12, window 0.025 s) for formants and autocorrelation (75–600 Hz range) for pitch, ensuring precise segmentation of 45 target words per participant against native speaker benchmarks. Qualitative data from interviews (n=24) highlighted increased learner engagement and confidence, with the "JEger–jeGER" mnemonic aiding stress placement. Observations confirmed VD's motivational appeal, aligning with Indonesia's educational policies. In conclusion, VD significantly boosts EFL learners' intelligibility through substantial WER reductions and prosodic gains, offering a scalable, technology-mediated model for pronunciation pedagogy that enhances communicative competence.