Abstract

This study proposes an intelligent framework that applies AI technologies throughout the entire process of collecting, managing, and utilizing oral history materials to enhance the efficiency and quality of oral documentation. Oral records consist of unstructured data, making them difficult to organize and search, and key procedures—such as transcription, verification, and content analysis—have therefore relied heavily on manual labor. To overcome these limitations, this study designs an integrated intelligent process that combines Speech-to-Text (STT) technology with Large Language Models (LLMs). Specifically, keyword boosting based on a domain-specific glossary was used to improve recognition accuracy for local proper nouns, and LLM-based automatic correction of punctuation and misrecognized words ensured consistency and accuracy in transcripts. In addition, generative AI was employed to automatically extract key terms, named entities, and summaries, and to structure this information to enable semantic linkages across oral records. Applying this model to oral history materials produced in an actual community documentation project demonstrated improvements in the automation efficiency of transcription and analysis, as well as more refined contextual understanding of oral content. This study shows that the paradigm of oral record management can shift from traditional manual “description” workflows to an “intelligent utilization-centered” model and provides foundational academic and practical insights for building intelligent oral archives in the future.