scieee AI-readable full text Open interactive document viewer

Captions and Transcripts: An AI Use Case

McFadden, Meredith

Abstract

Creating captions and transcripts for large quantities of audiovisual material is extremely time consuming. A-State has hundreds of hours of archival footage and oral histories that we want to make accessible. Our IR Team established a new workflow utilizing an LLM to generate transcripts, which are then reviewed.

Full text

An AI Use Case An AI Use Case Captions & Transcripts Captions & Transcripts by Meredith McFadden Arkansas State University Mid-South Center for Oral History THE PROBLEM Thousands of hours of film Hundreds of hours of oral history What A-State has... KAIT 8 News Footage Limited personnel Multiple large projects Judd Hill Oral Histories Bill Clements and Larry Ball Judd Hill Plantation CONSIDERATIONS Easy Editing Export Formats Regional Dialect Cost Platform access Training Material All platforms have an associated cost, most are time and/or subscription based Cloud or browser versus individual installation How well does the platform handle regional dialects? None are perfect but some are better than others How simple is it to edit the generated content? How intuitive is the platform? Does it allow all of the editing functions needed? Will your uploaded content be used to train their AI model? Will they keep a copy or reuse it in any way? Can we export the captions or transcripts? What formats are available? Step 4Step 3Step 2Step 1 Export from REV and burn captions in with Adobe Premier Create metadata and upload to ARCH Review by GA, student worker, and/or staff Run all materials through the AI to generate a base transcript or caption Editing based on Jackson State University Margaret Walker Center and Baylor University oral history style guides SITE CITE SIGHT Same sounding words are consistently the biggest failing of AI. The initial human review needs to be in full, then can go to spot checks. We’re going for “good enough” not perfect! SLURS & CURSES VOCABULARY CONSISTENCY I don’t know of any AI model that will include slurs and curse words, we have opted to retain them for accuracy of our historical records. We’ve added a disclaimer for these materials. Some platforms may have the option to build a unique vocabulary for regional, slang, or other uncommonly found words or phrases However you decide to present your captions and transcripts, be consistent. The AI model you choose will do a lot of the work here, but editors need to be aware of style guides in use. Be mindful of... A WORD ON EDITING CONCLUSION 1 Ask questions! Request a meeting with a sales person or customer service. A lot of need to know info isn’t readily provided. CONCLUSIONS Pick a few good clips to use as testers and run them through the various programs Free Trials or free basic subscription. Compare the results for accuracy, speed, etc. CONCLUSION 2 Set workflows for review and be prepared for a long-term project. AI cuts down a huge portion of work but review is still critical. CONCLUSION 3 THANK YOU! THANK YOU! Meredith McFadden, MLIS Institutional Repository Librarian [email protected] arch.astate.edu