This study explores obstruent characteristics in Korean dialects, particularly distinctions between lax, tense, and aspirated sounds, using phone-level speech recognition on a 2,200-hour corpus. Using the AI-Hub senior dialect speech dataset, we analyzed Gyeongsang (1,203 hours) and Jeolla (1,015 hours) dialects. A G2P model and wav2vec 2.0 XLS-R speech recognition model compared canonical with actual phone sequences. Gyeongsang dialect showed lax-tense confusion and lax-to-aspirated changes, while Jeolla dialect exhibited aspirated-to-lax changes. The lax-tense merger in Gyeongsang extends beyond previous /s/-/s*/ merger findings to general obstruents. Obstruent deletion in onset and coda positions in Gyeongsang represents a previously unreported phenomenon. The study demonstrates the effectiveness of large speech corpora for dialectal analysis, confirming known patterns while revealing new obstruent variations. Future research should examine how deletions vary by word position.
Lee et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: