Evaluating Collaborative Editing of AI-Generated Live Subtitles by Non-Professionals in German University Lectures
| Autor: | Piskorek, Patricia, Sienel, Nadine, Kuhn, Korbinian, Kersken, Verena, Zimmermann, Gottfried |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2022 |
| DOI: | 10.35011/icchp-aaate22-p1-22 |
| Popis: | ICCHP-AAATE 2022 Open Access Compendium "Assistive Technology, Accessibility and (e)Inclusion" Part I / Petz, Andrea ; Hoogerwerf, Evert-Jan ; Mavrou, Katerina, Seite 165-175 Subtitles are an important accessibility feature for deaf and hard of hearing students. The automatic generation of subtitles with artificial intelligence provides a fast and cost-effective way to make lectures more accessible. How-ever, automatically generated subtitles are not sufficiently accurate for compre-hension. One possible way to increase the accuracy of subtitles in a lecture setting is through a collaborative editing process by amateurs in a live setting. To this end, we conducted a Wizard-of-Oz test to simulate the process of live-editing a university lecture. Participants watched a lecture recording and collaboratively edited AI-generated subtitles in small teams. Subtitles were either presented sim-ultaneously or with a delay that simulated a realistic processing time of the AI. Participants reported their cognitive load and how well they could follow the content of the lecture. The accuracy of subtitles was measured using the Word Error Rate (WER). Manual editing under both conditions reduced the WER of the AI-generated subtitles. Participants receiving subtitles simultaneously achieved a lower WER, reported a lower cognitive workload and could follow the content of the lecture better than those experiencing a delay in the presenta-tion of subtitles. A minimal offset between the audio content and the presentation of subtitles supports amateurs in editing subtitles. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|