Empirical analysis of ASR error propagation in Korean SQA finds consistent relative downstream degradation across LLMs, single-character errors as a distinct semantic failure mode, and better performance from audio LLMs than matched ASR-LLM cascades.
An approach to measuring the performance of ASR models in the context of LLM-powered applications,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Analyzing Error Propagation in Korean Spoken QA with ASR-LLM Cascades
Empirical analysis of ASR error propagation in Korean SQA finds consistent relative downstream degradation across LLMs, single-character errors as a distinct semantic failure mode, and better performance from audio LLMs than matched ASR-LLM cascades.