Answers are grounded in this paper and cite their sources.
Background: This study addresses a gap in applied phonetics by developing an evidence-based, neural network-driven computer phonetic test for Russian EFL learners. Integrating interdisciplinary methods, the system targets Russian-specific pronunciation deviations and delivers adaptive feedback, thereby advancing both perception and production skills while aligning technological innovation with pedagogical effectiveness. Purpose: The purpose of the study is twofold: (1) to design and develop a computer-based system employing deep learning neural networks for objectively assessing Russian EFL students’ perception and production skills, and (2) to evaluate the effectiveness and reliability of this system through repeated testing, statistical analysis of learner performance and user feedback. Methods: A pre-test identified frequent segmental deviations, informing a targeted item pool. The software was developed in Microsoft Visual Studio 2022 (C#) using the Microsoft Speech Recognition Engine. The perception module used randomized audio stimuli (WAV files), while the speech recognition one recorded response via built-in microphones for automated accuracy evaluation. Twenty-five Russian EFL students (B1–B2 CEFR, aged 19–22) completed three test iterations at one-week intervals. Post-test questionnaire assessed usability and perceived learning gains. Data were analysed using descriptive statistics and correlation analysis. Results: We designed a computer-based system employing deep learning neural networks and assessed its efficiency in Russian EFL learners. The study found a 14.5% overall improvement in participant performance, with results showing a clear linear increase supported by a high R² value. Students performed better in perception tasks than in production practice. Pearson correlation analysis indicated consistent performance between consecutive attempts, supporting robust test-retest reliability. Both modules showed high internal consistency (α = 0.90 for perception, α = 0.88 for production). Participants rated the tool as useful and interesting, although they suggested improving the speech recognition function due to minor technical flaws. Conclusion: The module focused on testing perception skills can serve as an effective and engaging learning tool. While the pronunciation control component shows potential, its performance can be further enhanced through additional testing with high-sensitivity microphones to refine speech recognition accuracy. Overall, continued exploration of CAPT systems presents a promising direction for future research and innovation.
No takes yet. Share an insight, caveat, or question.
Kolesnichenko et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: