A mobile-web platform collecting atypical Korean speech across developmental and clinical populations

초록

We present a mobile-web integrated system for large-scale collection, processing, and analysis of atypical Korean speech. The system addresses the underrepresentation of children, elderly adults, and clinical populations in Korean automatic speech recognition (ASR) corpora and clinical resources. Our platform combines an Android recording application, implemented in React Native, with a Django REST Frame-work backend and a web dashboard for quality control and transcription. Population-specific interfaces are provided for children, older adults, and clinician-led sessions, and a flexible JSON-based metadata schema supports diverse clinical protocols while preserving longitudinal tracking of speakers. An integrated semi-automatic transcription pipeline uses Whisper/WhisperX models to generate draft transcripts and time alignment, which are then corrected and enriched by human annotators with dual orthographic-pronunciation transcriptions and disfluency labels. Automatic signal-quality checks and metadata validation prevent low-quality recordings from entering the corpus. The resulting infrastructure serves dual purposes: training Korean ASR models adapted to atypical speech and supporting remote or clinic-based speech assessment for developmental and age-related changes. We argue that language-specific, inclusive data collection frame-works such as this are essential for equitable access to speech technologies and clinically meaningful assessment tools in non-English contexts. © 2026 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution 4.0 International (CC BY) License (https://creativecommons.org/licenses/by/4.0/).