AI_LectureNote: A Retrospective Pilot Study of a Post-ASR Workflow for English-Script Rendering and Semantic Drift in Korean-English Medical Lectures

2026-07-19

AI Summary

This study addresses the degradation in readability and semantic fidelity caused by automatic speech recognition (ASR) systems transliterating Latin-derived medical terminology into Korean during Korean–English medical lectures. To enhance readability, the authors propose a post-processing workflow that reformats ASR output into structured lecture notes while restoring original English terms. The approach integrates outputs from Whisper-1 and GPT-4o-transcribe, employing chunked processing, term restoration, semantic consistency verification, and polarity cue analysis. The work reveals, for the first time, a decoupling between term restoration and semantic fidelity—demonstrating that surface-level accuracy does not guarantee semantic faithfulness—and introduces a multidimensional evaluation framework. Experiments show a substantial improvement in English term recovery (e.g., from 0.39 to 0.71 for Whisper), while uncovering 34–36 instances of semantic drift and 11–13 polarity errors, highlighting shared failure patterns across ASR frontends.

Abstract

AI_LectureNote is a historical, readability-oriented post-ASR workflow for Korean-English medical lectures. It rewrites speech-to-text output into study transcripts while restoring Latin-script medical terms rather than Korean phonetic transliterations. We retrospectively evaluate the workflow on four author-recorded lectures across five conditions. In this pilot, post-processing raised the macro English-script rendering rate from 0.39 to 0.71 on the whisper-1 path and from 0.26 to 0.65 when applied to 3-minute chunked gpt-4o-transcribe output. However, English-script rendering did not imply semantic faithfulness: the two post-processed conditions showed semantic drift in 34 and 36 of 282 reference sentences and polarity failures in 11 and 13 of 101 polarity-cue rows. A descriptive cross-input comparison suggested different candidate failure patterns: polarity-failure sets overlapped more strongly across front-ends (Jaccard 0.60; 9 shared of 15 unioned failures) than general semantic-drift sets (Jaccard 0.23; 13 shared of 57 unioned drifts). This single-annotator pilot documents concrete failure modes rather than population rates and supports evaluating surface accuracy, term-script rendering, chunk-level script consistency, and medical-meaning preservation separately.

Problem

• semantic drift
• English-script rendering
• post-ASR workflow
• medical lecture transcription
• polarity failure

Innovation

• post-ASR workflow
• English-script rendering
• semantic drift
• medical lecture transcription
• term restoration