The results of BUT teachers' ranking are out - congratulations to Mireia Diez Sánchez, Lukas Burget and Karel Beneš for scoring among the Top-10 at FIT in the Master program - it looks like machine learning courses are in good hands! In the Bachelor ranking, Jan "Honza" Černocký (the notorious signals and systems' torturer) scored the 7th. https://lnkd.in/eEcj9_XM
BUT Speech
Výzkumné služby
We do impactful research and raise new leading scientific personalities in the field of speech processing.
O nás
BUT Speech@FIT, founded in 1997, is one of the most famous speech data mining research and development groups in the world. Our mission is to do impactful research and raise new leading scientific personalities. The group is advised by Prof. Hermansky, managed by Dr. Jan "Honza" Cernocky, and its research director is Dr. Lukas Burget. BUT Speech@FIT has a significant track in EC-sponsored projects as well as funding by US Government, and local research agencies. Additionally, the group has extensive cooperation with international and local industrial partners and is active in open-source software development.
- Web
-
https://speech.fit.vutbr.cz/
Externí odkaz pro organizaci BUT Speech
- Obor
- Výzkumné služby
- Velikost společnosti
- 11 - 50 zaměstnanců
- Ústředí
- Brno
- Typ
- Vzdělávací společnost
- Datum založení
- 1997
Lokality
-
Primární
Božetěchova 2
Brno, CZ
Zaměstnanci společnosti BUT Speech
-
Josef Zizka
Co-Founder & CTO at ReplayWell
-
Jan "Honza" Černocký
Professor, Head of Department at Brno University of Technology
-
Martin Kocour
PhD student at Brno University of Technology, Speech@FIT | Researcher in speech recognition
-
Dominik K.
Speech Processing/ML Junior Researcher
Aktualizace
-
DiaPer: End-to-End Neural Diarization with Perceiver-Based Attractors by Federico Landini, Mireia Diez Sánchez, Themos Stafylakis, Lukáš Burget has been accepted by IEEE/ACM Transactions on Audio, Speech and Language Processing and is now available under the Early Access area https://lnkd.in/eZvcf7CU DiaPer proposes a fully-attentive alternative for computing speaker attractors in the context of end-to-end neural diarization. It reaches better performance than EEND-EDA with a more light-weight design, finding the number of speakers more accurately and better handling overlapped speech; all with a non-iterative single-stage system. You can also access our repository: https://lnkd.in/ecPiqZWG containing the code used for this work and models trained on free public data.
-
-
On Friday June 28th, the French embassy in Prague announced Joseph Fourier prizes for PhD research work in computer sciences. Karel Beneš from our group was awarded the third prize by H.E. Mr. Stéphane Crouzat - the French ambassador for Czechia. The 1st prize went to our faculty too, so we extend our congratulations also to Juraj Síč from the VeriFIT research group!
-
-
Last week, Johan Rohdin visited Eleni Sergidou and her colleagues at Netherlands Forensics Institute (NFI). The visit was part of a research collaboration on speaker recognition based on linguistic information instead of (or in addition to) the acoustic information we normally use. The collaboration started during the EU ROXANNE Project 2019-2022.
-
-
For a short period of time, we had everyone in Brno to celebrate the Best Paper Award we obtained in Odyssey a few days ago! We thank the organizers for the recognition and for an excellent workshop https://lnkd.in/egDf8TNn If you still have not read the paper, see the link below. Do End-to-End Neural Diarization Attractors Need to Encode Speaker Characteristic Information? by Lin Zhang, Themos Stafylakis, Federico Landini, Mireia Diez Sánchez, Anna Silnova and Lukáš Burget Paper: https://lnkd.in/eXPvgkAq GitHub: https://lnkd.in/eaPH4PKb
-
-
During last year's JSALT workshop, we saw that multicultural teams of researchers from established companies, universities, and startups can deliver incredible results. Therefore, we decided to meet again this year (in Brno) and work on several conversational AI topics - optimal dialogue representation, an outlier/novelty detection in topic modeling, shared (text speech) representation models for dialogs, and open-source dialog system runtime. If you are around, stop by for a discussion. The participants are from BUT Speech, Idiap Research Institute, Omilia - Conversational Intelligence, Phonexia, Salted CX, Telefónica research and University of Novi Sad. Thanks FIT BUT, RedHat Brno, Phonexia, and Salted.CX for support.
-
Federico Landini will defend his PhD thesis “From Modular to End-to-End Speaker Diarization” next Thursday 27 June at 10:30 in meeting room G108. His supervisors were Lukáš Burget and Mireia Diez Sánchez. Federico graduated from University of Buenos Aires and joined BUT in 2017. During his PhD, he became a true “diarization celebrity” – his paper on variational Bayesian hidden Markov model approach for diarization published in Computer Speech & Language in 2021 gathered already 176 citations, and the VBx toolkit he maintains is a de-facto standard in the community. During his PhD, he also did several internships in good international labs (Facebook/Meta and Apple). We are grateful to Sriram Ganapathy (IISC, Bengaluru) and Hervé Bredin (CNRS IRIT, Toulouse) for accepting to be the reviewers of this thesis. https://lnkd.in/eaWZzFBa
-
We cordially invite you to the VGS-IT lecture "Factorized self-supervision models for speech representation learning" of Sriram Ganapathy from IISC Bengaluru - next Wednesday 26th June in room E112. Sriram was one of the Survey speakers at Interspeech 2021, but could not make it to Brno due to COVID restrictions, we are happy to have him finally visiting in person! https://lnkd.in/gRk3_8Tt
VGS Invited Talks @ FIT
vgs-it.fit.vutbr.cz
-
Today in Odyssey in the speaker diarization session, Fede will be presenting Lin's paper: Do End-to-End Neural Diarization Attractors Need to Encode Speaker Characteristic Information? by Lin Zhang, Themos Stafylakis, Federico Landini, Mireia Diez Sánchez, Anna Silnova, Lukáš Burget Paper: https://lnkd.in/eJgT46uC Repository: https://lnkd.in/eaPH4PKb
GitHub - BUTSpeechFIT/EENDEDA_VIB
github.com
-
Bolaji has successfully defended his PhD thesis "End-to-End Open Vocabulary Keyword Search" on Monday 27th May. Although member of the BUT group for several years, Bolaji was enrolled at the Boğaziçi University in Istanbul, Turkey, and his PhD work was supervised by Prof. Murat Saraclar. Once the thesis becomes publicly available, have a look at it - it is a very informative reading about neural embedding-based techniques for keyword spotting. Congratulations to our new doctor!
-