Biography:

Tomoki Toda was born in Aichi, Japan on January 18, 1977. He earned his B.E. degree from Nagoya University, Aichi, Japan, in 1999 and his M.E. and D.E. degrees from the Graduate School of Information Science, NAIST, Nara, Japan, in 2001 and 2003, respectively.

He is a Professor at the Information Technology Center, Nagoya University. He has also been a Visiting Researcher at the NICT, Kyoto, Japan, since 2006. He was a Research Fellow of JSPS in the Graduate School of Engineering, Nagoya Institute of Technology, Aichi, Japan, from 2003 to 2005. He was then an Assistant Professor (2005-2011) and an Associate Professor (2011-2015) at the Graduate School of Information Science, NAIST. From 2001 to 2003, he was an Intern Researcher at the ATR Spoken Language Communication Research Laboratories, Kyoto, Japan, and then he was a Visiting Researcher at the ATR until 2006. He was also a Visiting Researcher at the Language Technologies Institute, CMU, Pittsburgh, USA, from October 2003 to September 2004 and at the Department of Engineering, University of Cambridge, Cambridge, UK, from March to August 2008. His research interests include statistical approaches to speech, music, and sound information processing.

He received more than 10 paper awards including the 18th TELECOM System Technology Award for Students and the 23rd TELECOM System Technology Award from the TAF, the 2007 ISS Best Paper Award from the IEICE, the 2009 Young Author Best Paper Award from the IEEE SPS, and the 2013 Best Paper Award (Speech Communication Journal) from EURASIP-ISCA. He also received the 10th Ericsson Young Scientist Award from Nippon Ericsson K.K., the 4th Itakura Prize Innovative Young Researcher Award from the ASJ, the 2012 Kiyasu Special Industrial Achievement Award from the IPSJ, and the Commendation for Science and Technology by the Minister of Education, Culture, Sports, Science and Technology, the Young Scientists' Prize in 2015. He served as a member of the Speech and Language Technical Committee of the IEEE SPS from 2007 to 2009 and 2014 to 2016. He has served as an Associate Editor of IEEE Signal Processing Letters since Nov. 2016. He is a member of IEEE, ISCA, IEICE, IPSJ, and ASJ.

Title:

Recent Trend of Voice Conversion Research and Its Possible Future Direction

Abstract:

Voice conversion is a technique for modifying speech waveforms to

convert non-/paralinguistic information into any form we want while

preserving linguistic content. It has been dramatically improved thanks

to significant progress in machine learning techniques, such as deep

learning, as well as significant efforts to develop freely available

resources. In this talk, I will review recent progress of voice

conversion techniques, overviewing recent research activities including

Voice Conversion Challenges, and then, I will also discuss possible

future directions of voice conversion research.

Biography:

Prof. Hiroyuki Shinnou worked as a researcher in Fuji Xerox Co., Ltd. and Panasonic Corporation during 1987 and 1993,

He joined the Faculty of Engineering, Ibaraki University in 1993, as a research assistant.

After receiving his Ph.D. degree in Tokyo Institute of Technology in 1997, he worked as a lecturer, and an associate professor in Ibaraki University, respectively. He is currently a professor at the Department of Computer and Information Sciences in Ibaraki University.

Prof. Shinnou has long been active in the academic associations related to natural language processing, including ACL (Association of Computational Linguistics), JSAI, and IPSJ. Now, he serves as the director of the Association for Natural Language Processing (ANLP), and serves as the conference chairman in the annual conference NLP 2020 this year, which is the most important conference on Natural Language Processing in Japan.

Prof. Shinnou has published many academic papers in international journals such as ACM TALIP, and Natural Language Processing (in Japanese), and international conferences including ACL, PACLIC, LREC.

His research interests include Bayes statistics, machine learning, natural language processing and image processing. Since he integrates theory with practice, he also published many books (mostly in Japanese), which have great impact in related fields. Recently, he is actively researching deep learning technology, especially transfer learning.

Title:

Use of BERT for NLP tasks by HuggingFace's transformers

Abstract:

The pre-trained BERT model has been improving states of many NLP tasks.

I believe that the use of BERT is essential when we build some kind of

NLP system in the future. Initially, it was hard to use BERT because the

concept of the pre-trained model was unfamiliar, and BERT was available

only by using TensorFlow which is cumbersome for beginners. However,

today, there is the HuggingFace's transformers library. Thanks to this

library, everyone can utilize BERT easily.

In this talk, first I will explain what BERT is and what we can do by BERT,

and then I show some examples of the use of BERT by HuggingFace's

transformers. As an application, I will do fine-tuning of BERT for a

document classification task. Additionally, I will show the technique to

learn just some of the layers in BERT. As one of the improvements of

BERT, the study on smaller BERT model have been active, for example,

Q8BERT, ALBERT, DistilBERT, TinyBERT and so on. Even simple pruning of

BERT is effective. I will introduce these studies and show that some of these

models are available through HuggingFace's transformers.