Community

Organization of events

"Speaker recognition summer 2010" at FIT BUT, Brno

KALDI 2010 - workshop devoted to KALDI toolkit development, August 2010.

ICASSP 2011 - IEEE International Conference on Acoustics, Speech and Signal Processing, proceedings, recordings on Superlectures. Co-organized with Petr Tichavsky (UTIA Prague) and Ales Prochazka (Institute of Chemical Technology, Prague).

KALDI 2011 - workshop devoted to KALDI toolkit development, August 2011.

KALDI 2012 - workshop devoted to KALDI toolkit development, August 2012.

BOSARIS 2012, 4-week research workshop from October 29 until November 26, 2012.

KALDI 2013 - workshop devoted to KALDI toolkit development, August 2013.

ASRU 2013 - IEEE Workshop on Automatic Speech Recognition and Understanding, December 8-12, 2013, Olomouc. Proceedings, recordings on Superlectures.

INTERSPEECH 2021 - the theme was "Speech everywhere", 30 August - 3 September 2021 in Brno.

Participation and leadership

2007 JHU workshop group "Recovery from Model Inconsistency in Multilingual Speech Recognition" headed by Hynek Hermansky, with Lukas Burget, Petr Schwarz and Pavel Matejka in the team. This workshop advanced the detection and dealing with out-of-vocabulary (OOV) word detection and handling, that significantly helped us in the EU FP7 DIRAC project.

2008 JHU workshop group "Robust Speaker Recognition Over Varying Channels", with Lukas Burget, Ondra Glembek and Valia Hubeika in the team. The workshop gave birth to iVectors (the dominating paradigm in speaker recognition nowadays). After the workshop, Ondra Glembek came with techniques to simplify and optimize iVector extraction which brought him the Ganesh Ramaswamy prize at ICASSP 2011 for his paper"Simplification and Optimization of i-vector extraction".

2009 JHU workshop group Low Development Cost, High Quality Speech Recognition for New Languages and Domains, with Lukas Burget and Petr Schwarz on the team. Besides work on sub-space GMMs, this workshop was at the origin of the nowadays extremely popular KALDI toolkit. Several KALDI workshops followed in Brno, see above.

2013 JHU workshop group Speaker and Language Recognition with Lukas Burget, Oldrich Plchot and Karel Vesely on board. This workshop worked hard on adaptation of speaker recognition to new domains, and was marked by Domain Adaptation (or simply Doug's) challenge.

2014 1st Frederick Jelinek Memorial Workshop (Prague) group "ASR Machines That Know When They Do Not Know" with Hynek Hermansky, Lukas Burget and Karel Vesely on board. The main topic investigated by BUT guys were confidence measures allowing to say whether to trust the output of acoustic modeling DNN.

2015 Speaker Recognition Summer Workshop (Torino) with Lukáš Burget, Oldřich Plchot and Ondřej Novotný in the team. The workshop dealt with multiple SRE topics, including adaptation and text-dependency.

2015 2nd Frederick Jelinek Memorial Workshop (Seattle) group "Far-Field Enhancement and Recognition in Mismatched Settings" with Martin Karafiat, Karel Vesely, Katka Zmolikova and (partly) Honza Cernocky on board. BUT people concentrated on data augmentation and DNN-based methods for data selection and ASR domain adaptation.

2015 QUESST task in MediaEval - BUT (Igor Szoke) was main organizer of QUESST task aiming at Query-by-Example search in speech. 10 labs from the whole world participated.

Invited talks

Honza Cernocky: Word-subword based keyword spotting with implications in OOV detection, ASILOMAR, Pacific Grove, November 2010.

Lukas Burget: Recognizing a Million Voices: Low Dimensional Audio Representations for Speaker Identification, Microsoft Research, March 2nd 2011, video.

Lukas Burget: iVectors: Low Dimensional Representations of Speech for Speaker Verification and Related Fields, Google, Mountain View, September 19th 2011.

Honza Cernocky: Building speech recognizers with limited resources, Iberspeech 2012, 21 November 2012, Madrid, Spain.

Igor Szoke: Speech-to-text and search in speech, Telefonica R&D, Barcelona, April 5, 2013.

Frantisek Grezl: From Probabilistic to Stacked Bottle-Neck features, UT Dallas, May 2014.

Lukas Burget: Subspace Modeling Techniques in Speech and Language Processing, IEEE SLT 2014, 9 December 2014, South Lake Tahoe, USA.

Honza Cernocky: Ověřování osob pomocí hlasu (Speaker identification by voice - in Czech), AEC Security conference, Prague, 18 February 2015.

Honza Cernocky: Dolování dat z řeči pro bezpečnostní aplikace (Speech data mining for security applications), Security Session, Brno, 11 April 2015.

Ondrej Glembek: Voice-print transformation for migration between automatic speaker identification systems, 7th European Academy of Forensic Science Conference, Prague, September 6-11 2015.