Extraordinary closed captions: correcting academic language - with Nelly Iacobescu & Martina Emmerich
At the University of Edinburgh, we record and automatically transcribe our lectures and community-created media. Student interns correct transcripts where Automatic Speech Recognition (ASR) fails to detect technical terms, names, or struggles with our many international accents. With increasing media production, we are streamlining this process. Using our dataset of highly accurate transcripts, lecture slides, and ASR outputs, we are prompting LLMs to look for context, identify and correct errors for inclusive educational resources.
Key takeaways
How to supporting learning by offering accurate materials to students and the community.
Improving Automatic Speech Recognition with AI
Use technology to improve the accessibility of content for differently abled and non-native English learners.
About your speaker
Nelly Iacobescu manages the Captioning Service and AI components of Media Services at the University of Edinburgh. Following Masters’ degrees in Audiovisual Communication and Film Studies, Nelly has over 10 years’ experience in film editing and media content creation.
Martina Emmerich is a Research Assistant in Automatic Speech Recognition (ASR) at the Centre for Speech Technology Research (CSTR), University of Edinburgh. With an MSc in Speech and Language Technologies, Martina specialises in speech technology and natural language processing. Her work focuses on enhancing lecture caption accuracy for domain-specific terminology by leveraging Large Language Models (LLMs) and biasing ASR model outputs.