Developing Speaking Skills with Neural Network Technologies: An Experimental Study

Abstract

Background. In recent years, the classroom hours allocated to foreign language teaching in higher education have decreased significantly, hindering the development of communicative competence — a key component of professional training, even in technical fields. This shortage necessitates innovative solutions to enhance speaking skills development outside traditional classroom settings. Multimodal AI, capable of analyzing and contextually responding to speech, offers a viable solution by enabling interactive practice in both written and oral communication. Its adaptability makes it a highly promising tool for modern language education.

Objective. This study aims to assess the efficiency of neural network-based tools for improving speaking skills in foreign languages.

Study Participants. The study involved 84 participants, all students from the Institute of IT and Cybernetics at Samara University.

Methods. Theoretical Methods: comparative analysis of scientific publications relevant to the research topic. Empirical Methods: a formative experiment with control and experimental groups to assess the development of speaking skills. Quantitative data were analyzed using statistic methods: Shapiro — Wilk test, Mann — Whitney U-test, one-way ANOVA with effect size measured by omega-squared (ω²), and Wilcoxon signed-rank test.

Results. An assessment framework was created to evaluate the development of speaking skills. The experiment lasted 18 months. Data analysis showed statistically significant improvements (p < 0.001) across all target metrics due to neural network implementation. The strongest effect was revealed in speech fluency.

Conclusions. The findings confirm that neural network technologies significantly enhance speaking skills development.