Audio Description in the Era of AI: Challenges and Future Development

Audio Description (AD) provides accessibility service to people with visual impairment. AD has received increasing attention from both the public and academia over the past 20 years, as a result, research and practice in this field have developed rapidly. Given the high degree of similarity between the processes of translating between languages and translating visual content into words, it is possible to apply the principles of translation studies (TS) to AD, consequently, AD has been considered a form a translation (audio-visual translation or intersemiotic translation). Just like the industry of translation, AD is facing the challenges brought by the recent development in AI. For example, technological innovations such as ChatGPT, automatic speech recognition, machine translation have dramatically changed the prospect of translation and AD studies. This study, based on a review of AD-related technological development and an experiment comparing AD output generated by human professionals and AI-powered tools, attempts to examine the challenges and possible direction for future development in this field. A sample of video slips is chosen for the production of both human-created and AI-generated AD transcripts, which will be evaluated by a group of 4 researchers consisting of both sighted and visually impaired people. Factors such as choice of details, language use, contextual appropriateness, and overall user satisfaction will be discussed. The results will provide insights into the strengths and limitations of human-generated and AI-based AD, and will shed light on how AI-powered AD may complement or augment human efforts, focusing on opportunities for collaboration and workflow optimization.