/
Personal Bio

Kris Demuynck received the master degree in Electrical Engineering and the PhD degree in Applied Sciences from the Katholieke Universiteit Leuven, in 1993 and 2001, respectively. Between 2001 and 2012 he worked as a Postdoctoral Researcher at the Katholieke Universiteit Leuven on various national and international projects. In 2012 he joined the Multimedia Lab (now Data Science Lab) at Ghent University, department of Electronics and Information Systems (ELIS), where he works as part-time professor and research manager. His principal research interests are large vocabulary continuous speech recognition, machine learning and search algorithms. His PhD work and the subsequent research resulted, amongst others, in a speech recognition toolkit which was made available publicly under the name SPRAAK in 2008. Some of the speech & language technology related topics on which Kris Demuynck has worked are: search algorithms (frame synchronuous FST-based decoding, asynchronous lattice decoders, efficient k-NN algorithms, ...), algorithms to extract information from audio (prosody, speaker identification, noise tracking, ...), deep learning (with MLPs, CRFs and reservoir computing networks), various forms of language modeling (N-grams, topic models, cache models, recurrent neural networks), and various non-standard (human-inspired) approaches to speech recognition.

Presentation title: 
AI & Speech Technology -- past, present & future
Presentation description: 
In this short talk, I'll first explain what automatic speech recognition encompasses and why speech recognition is so difficult for machines. In this part of the talk, I'll also cover the knowledge that must be incorporated in the system in order to get a machine to understand speech, and I'll explain how all information sources are combined to find the most plausible solution. In the second half of the talk, I'll look at how deep learning improves automatic speech recognition, and at what is still needed to attain near human performance and a human-alike behavior in the future.