A joint event between ELOQUENCE and the AI-SPEAK project at the Faculty of Technical Sciences, University of Novi Sad
On Saturday, 25 April, the Science and Technology Park in Novi Sad hosted the final event of the AI-SPEAK competition, a hackathon organised in partnership with ELOQUENCE and the Faculty of Technical Sciences (FTN), University of Novi Sad (UNS). The event drew more than fifty attendees and showcased some of the most exciting student work in audio-visual speech synthesis and multimodal AI coming out of the region.
A shared mission
The collaboration between ELOQUENCE and AI-SPEAK is a natural one. Both initiatives are working on the frontiers of multilingual, multimodal voice technology, building systems that can listen, speak, and now also look convincingly human. By joining forces for this hackathon, the two projects gave students a chance to tackle a problem that sits right at the intersection of their research agendas: generating an animated virtual avatar whose facial movements convincingly follow synthesised speech.
It is exactly the kind of challenge that requires expertise from across disciplines, signal processing, machine learning, computer animation, and human-computer interaction, and the participating teams reflected that breadth.
What the teams worked on
The three finalist teams each tackled the core competition task: text-driven audio-visual speech synthesis, where a virtual character must be animated so that its lip and facial movements match a synthesised voice in a natural way.
Alongside the main competition, six showcase (“revijalni”) projects were presented by students from a wide range of study programmes at FTN, including:
– BSc Animation in Engineering
– BSc Information Engineering
– BSc Biomedical Engineering
– MSc Energy, Electronics and Telecommunications (Signal Processing module)
– MSc Artificial Intelligence and Machine Learning
The showcase projects explored adjacent problems such as automatic lip-reading in Serbian and language identification from lip movements, topics that are directly relevant to ELOQUENCE’s broader work on inclusive, robust speech technology.
Three approaches, one problem
What made the final particularly interesting was watching three quite different engineering philosophies converge on the same task.
One team achieved the best visual result by pairing a relatively simple base model with strong post-processing, a reminder that clever pipeline design can rival raw model complexity. A second team went the other way, using state-of-the-art models to produce very natural animation at the cost of a heavier, more complex system. The third team delivered the methodologically cleanest solution, with a particularly thoughtful approach to real-time performance and loss-function design, even if the final visuals were slightly less polished.
Taken together, the three projects offered a clear illustration of the central trade-off in this field: visual quality, system complexity, and methodological rigour rarely come for free, and good engineering is about choosing where to spend your budget.
The jury evaluated the projects on multiple criteria: animation quality, real-time performance, the quality of the written report, innovation in approach, and the presentation itself, including how the teams handled questions from the expert panel.
The results
1st place: Milica Bošnjak, Nikola Stojičić, Tijana Petrović and Iva Jovanović
MSc Energy, Electronics and Telecommunications, Signal Processing module
2nd place: Iva Marković, Neda Šerović, Dragomir Božoki and Andrej Dušanić
MSc Energy, Electronics and Telecommunications, Signal Processing module
3rd place: Olivera Miličić and Iva Bojanić
BSc Animation in Engineering
Best showcase project: Vasilije Šolaja
BSc Animation in Engineering
Congratulations to all the winners, and to every student who took part. The quality of the work on display made the jury’s job a difficult one.
Why this matters for ELOQUENCE
For ELOQUENCE, supporting events like this is not just about spotting talent (although there was plenty of that on the day). It is about strengthening the European ecosystem around multimodal, multilingual voice AI, and about making sure that languages like Serbian, which are sometimes underserved by mainstream speech technology, are represented in the next generation of tools and the next generation of researchers building them.
The AI-SPEAK competition is a great example of how a focused national project can open up onto a broader international community. We are proud to have been part of it, and we are looking forward to seeing where these students take their work next.
