When Multiple Scripts Matter: Evaluating ASR in Clinical Settings

The article introduces MultiClin, a benchmark for evaluating automatic speech recognition (ASR) systems in clinical settings that addresses multiscript variability. It reveals that traditional evaluation metrics underestimate performance by not accounting for orthographic variants, and demonstrates that multiscript-aware evaluation provides a more accurate assessment. Key findings indicate that inconsistent script mappings hinder model convergence, while script unification leads to optimal ASR performance, highlighting important considerations for practitioners working with ASR in diverse linguistic environments.