Automated speech analysis using machine learning methods for language competencies assessment in Russian as a foreign language exam

The article explores the application of computer and digital technologies in assessing students’ speech and communication skills in the Russian as a Foreign Language (RFL) exam, particularly in the “Speaking” subtest. The study analyzes existing approaches to automated speech assessment, identifying their advantages and limitations. Based on empirical research conducted on a sample of 120 students in 2022–2023, the effectiveness of using computer algorithms for analyzing prosodic, lexical, and grammatical characteristics of examinees’ speech is demonstrated. The application of machine learning methods allowed achieving a high correlation (r = 0,87; p < 0,01) between expert assessments and automated analysis results. Particular attention is paid to the creation of a three-factor model that takes into account speech fluency, lexical-grammatical and phonetic-prosodic characteristics. This model demonstrates high prediction accuracy (R2 = 0.79) and has successfully undergone cross-validation on an independent sample. The data obtained indicate the promising potential of integrating computer technologies into the system of speech skills assessment while maintaining the key role of expert evaluation. The research results are significant for improving RFL teaching methods and developing tools for objective assessment of communicative competence.