Our Themos Stafylakis, Head of Machine Learning and Voice Biometrics at Omilia, attended Interspeech where he presented “Speaker embeddings by modeling channel-wise correlations”.

Interspeech21 is the world‘s largest technical conference focused on the science and technology of spoken language processing, hybrid event, physically in Brno, Czech Republic, and also virtually on INTERSPEECH 2021.

As per Themos words, “Interspeech2021 in Brno (Czech Republic) was the first conference I physically attained after two years of visual-only “conferences”. In fact, the previous one was Interspeech2019 in Graz (Austria) and I travelled there from Brno by train.
I am a frequent visitor of Brno, thanks to the collaboration and friendship I have with the speech team of the Brno University of Technology (BUT). It started in 2010, when they organized the “Odyssey” Workshop on Speaker Recognition and I went there as a PhD student to present my paper on speaker diarization. BUT is a top-tier player in speech technologies, witnessed also by the fact that they are the organizers of some of the top conferences in speech.
I also had the chance to present the paper “Speaker embeddings by modeling channel-wise correlations”, the latest outcome of the collaboration between Omilia and BUT. In this paper, we demonstrated how style-transfer methods in computer vision can be applied to capture speaker characteristics. The overwhelming interest in our work from the speaker recognition community indicates that it will have an impact on the field.
Whether we like it or not, conferences are valuable to researchers and the physical presence and interaction with the community cannot be (fully) replaced by virtual ones. This was the consensus between all ~300 researchers who attended Interspeech2021 physically, and I believe between the other >2000 ones who attended it virtually.
So, well done BUT for organizing such a successful event and let’s keep on doing great work together."

To check presentation from the event