BUT Speech@FIT is part of the Department of Computer Graphics and Multimedia at the Faculty of Information Technology, Brno University of Technology. It is led by Prof. Jan “Honza” Černocký, Prof. Lukáš Burget and Prof. Hynek Heřmanský (also Emeritus at the Johns Hopkins University). It includes 30 faculty, researchers, PhD and pre-graduate students and support staff. BUT SPeech@FIT crew is international and, besides expected Czechs and Slovaks, includes 7 nationalities (Sweden, Russia, Spain, China, India, Argentina, and Nigeria).

BUT Speech@FIT deals with artificial intelligence and machine learning applied to speech data mining: speaker and language identification, speech recognition, and keyword spotting. The group was a key contributor to wide-spread use of neural architectures in speech and language processing (Tomas Mikolov, the founder of RNN language models and word2vec, is our PhD graduate). In recent years, the scope was enlarged towards integration with natural language processing (NLP), automated dialog systems, and voice deep-fake detection. The group extensively publishes and provides most of its research code as open-source.

The group has significant track of research projects - more than 20 European ones (FP 4-7, Horizon 2020, Horizon Europe), and has been funded by U.S. DARPA, IARPA, EOARD, ONR.and almost all Czech funding agencies.

Highlights:

• In 2021, Aminer’s AI 2000 Most Influential Scholars analysis ranked BUT among 5 Speech recognition most influential organizations.
• In the same ranking, 4 group members and former PhD students (Lukas Burget, Jan Cernocky, Tomas Mikolov, and Pavel Matejka) were among the 100 most cited scholars in the area of speech recognition.
• Strong presence in prestigious speech/NLP research workshops (Frederick Jelinek Memorial Summer Workshop - JSALT) organized by Johns Hopkins University – team leaders, etc. 2025 JSALT will take place at FIT BUT.
• Significant experience with large models AI models - BUT is coordinator of an EU contract aimed at EC-owned large ASR model.
• Track in several US projects (funded by EOARD, DARPA, IARPA).
• Key member of a consortium selected to run speech and NLP evaluations in the European Defence Fund (EDF) HLTO call.
• Key contributor to the community, organizer of events and workshops, including Interspeech (the largest speech conference) in 2021 in Brno.
• Three spin-off companies, of which Phonexia has more than 40 FTE and is selling production speech technologies to more than 60 countries world-wide.

The group is the primary user of FIT’s computing cluster with more than 1500 CPU cores in several blade chassis, file servers with total capacity of more than 1 PetaByte, and more than 100 GP-GPU cards for scientific computation. As FIT BUT was the founding partner of the IT4Innovations Center of Excellence), we have a priority access to the Karolina supercomputer and the top European HPC system LUMI in Finland.