11am - 12 noon

Thursday 12 July 2018

Blind source separation based on unsupervised and semi-supervised learning for multichannel audio data

Professor Hiroshi Saruwatari from the University of Tokyo, Japan will be speaking.


University of Surrey
back to all events

This event has passed


Blind source separation (BSS) is an unsupervised learning approach for estimating original source signals using only mixed signals observed in multichannel inputs. In particular, BSS algorithms based on independent component analysis (ICA) and independent vector analysis (IVA), in which the independence among source signals is mainly used for the separation, have been studied actively in the past decade.

In this lecture, looking back their history from ICA to IVA, we focus our attention on the new extension to low-rank spectrogram modeling and sparse representation, and introduce independent low-rank matrix analysis (ILRMA). In ILRMA, several source models based on complex heavy-tailed distributions are explained with the discussion on fruitful relation between non-Gaussianity and low-rankness.

Finally, thanks to audio big data capability, ILRMA and deep learning are combined, resulting in the sophisticated hybrid method “independent deeply learned matrix analysis (IDLMA).” In addition to the theoretical basis of the algorithms, some applications combining BSS and real-world audio system will be reviewed, e.g., binaural hearing-aid system and distributed microphone array system for speech detection.


Hiroshi Saruwatari received the B.E., M.E., and Ph.D. degrees from Nagoya University, Japan, in 1991, 1993, and 2000, respectively. He joined SECOM IS Laboratory, Japan, in 1993, and Nara Institute of Science and Technology, Japan, in 2000. From 2014, he is currently a Professor of The University of Tokyo, Japan. His research interests include statistical speech signal processing, blind source separation (BSS), audio enhancement, and robot audition. He has successfully achieved his carrier, especially on BSS researches including theoretical bridge between unsupervised learning and spatial signal processing, and development of the real-time algorithm.

He has put his research into the world's first commercially available Independent-Component-Analysis-based BSS microphone in 2007. He published 95 refereed original papers of international journals and 330 conference papers, getting more than 5700 citations. He received paper awards from IEICE in 2001 and 2006, from TAF in 2004, 2009 and 2012, from IEEE-IROS2005 in 2006, and from APSIPA in 2013. He received DOCOMO Mobile Science Award in 2011, Ichimura Award in 2013, The Commendation for Science and Technology by the Minister of Education in 2015, and Achievement Award from IEICE in 2017. He won the first prize in IEEE MLSP2007 BSS Competition. He has been professionally involved in various volunteer works for IEEE, EURASIP, IEICE, and ASJ, including chair posts of international conferences and associate editor of journals.

Visitor information

Find out how to get to the University, make your way around campus and see what you can do when you get here.