This study investigates the potential and limitations of Google Gemini in assessing German pronunciation. By bypassing conventional STT methods and processing audio waveforms directly, it examines Gemini’s ability to analyze critical (supra-)segmental features such as vowel length, diphthongs, and stress realization. Specifically, the research quantitatively evaluates Korean learners’ pronunciation to identify Gemini’s analytical boundaries and recognition errors from a phonological perspective. These findings provide essential empirical evidence and foundational data for the future implementation of AI-driven automated scoring systems in German language education.
Se-Jin Park (Tue,) studied this question.