On February 6, 2026, Róbert Fjólnir Birkisson, an expert in the application and implementation of AI solutions, will give a lecture on the next generation of speech-to-text recognition: From simple speech recognition to multimodal models. In his talk, Róbert will discuss how speech recognition models for Icelandic are trained on audio and corresponding text, thus learning to convert speech into text. Post-processing the output ensures increased readability and quality. Although traditional models achieve good results, the new generation of models opens up the possibility of training them directly on texts to develop a built-in linguistic intuition. Such training leads to better Icelandic, fewer errors, an improved user experience, and opens up possibilities for more diverse output, for example, identifying who is speaking and when, or creating more detailed and varied audio processing.

Róbert has extensive experience in bridging the gap between complex computational models and practical software to create a better user experience for users and companies. He has worked on the development and implementation of AI solutions at Miðeind for the past year and a half. Róbert holds an MSc in Complex Adaptive Systems from Chalmers University of Technology in Sweden and a BSc in Software Engineering from the University of Iceland.

UTmessan, held at Harpa, is one of the largest technology events of the year in Iceland. It brings together a diverse range of companies, experts, and technology enthusiasts to share knowledge, discuss the latest trends in the tech world, and spark interest in computer and technology fields. Róbert's lecture is ideal for anyone who wants to gain a deeper understanding of speech recognition and how the latest AI models work.

We encourage everyone in the area to attend and listen to an informative lecture that connects science, technology, and practical application in an exciting and accessible way. You can view the schedule here.