Institutional development · Digital transformation, data and AI · Germany
TH Deggendorf develops AI system that reads lecture slides to subtitle videos
Technische Hochschule Deggendorf reports that its TACT system reads on-screen slides to train an AI for subtitling recorded lectures, and claims it beats large IT companies' systems using minimal computing power.
Why it matters: Universities that publish lecture videos face a subtitling duty the announcement says is hard to meet; THD reports a system it says handles dialect and jargon cheaply.
A system called TACT for subtitling lecture videos has been developed at the Technische Hochschule Deggendorf (THD), according to the announcement.
According to THD, TACT reads the slides shown in a video and uses them to train a language-capable AI, so that it copes with dialect, complicated technical terms and poor recordings.
THD claims the system outperforms those of large IT companies while needing only minimal computing power.
The announcement states that universities have been legally required to provide such subtitles for years, but that in practice this always founders on the effort and technology involved.
Editor’s view
Sarah Williamson, Teaching, Digital and AI Editor
The announcement identifies universities obliged to subtitle lecture videos as the group held back by effort and technology, and THD reports that TACT tackles accuracy by learning from lecture slides while using minimal computing power.
