University of Minnesota
http://twin-cities.umn.edu/
612-625-5000

Home

Speech technologies at Google: An overview

by

Giorgio Quer
Director of Artificial Intelligence
Scripps Research Translational Institute
San Diego, CA

Friday, October 18, 2019
3:00–4:30 pm

101 Walter Library

IEEE SPS Society Distinguished Lecture Series

NOTE: Registration (required)

Please register by emailing Sandy Jobes your names and affiliation at Sandy_Jobes@starkey.com so that we can plan for the conference room, snacks and beverage in advance.

In this talk I'll give an overview of google speech technologies with a bigger focus on speech recognition. I'll describe the genesis of speech activities at google, its evolution over the last 15 years, and where speech technologies are moving in the future. I will then spend a bit more time on the current state of the art in neural networks for speech modeling: sequence to sequence neural nets. Sequence to sequence modeling bring some interesting advantages and simplifications in speech recognition, synthesis and voice conversion. Tasks that before were hard like multi-linguality are much simplified now. I'll finish the talk describing our new work to help users with dysarthric speech, PARROTRON.

 

Giorgio Quer joined Scripps Research Translational Institute in January 2017. His expertise is in artificial intelligence and probabilistic modeling applied to heterogeneous data signals, in order to extract key information and make predictions on future occurrences based on past data. His contributions include new methods exploiting compressive sensing to collect and process wireless sensor network data, data link layers protocols for cognitive networks, and new ways to extract information from heart rate variability in order to study group dynamics.

Quer’s research interests are focused on the interpretation and representation of big data for human health, in order to build models for prediction of future occurrences and improve patient outcomes. His multi-disciplinary interests include theoretical models, such as compressive sensing, Bayesian analysis, wavelet coherence and Markov decision processes; and analysis of noisy time-series from wearables, in particular physiological signals such as blood pressure, heart rate variability and photoplethysmography.

At the Translational Institute he works on the data analytic side of the All of Us Research Program, adopting probabilistic models and predictive analytics to extract information from large health datasets available through the program, as well as from other industrial collaborations. His goal is to extract and present this information in a useful way to clinicians and other users.

Quer received his undergraduate and master's degrees with honors in Telecommunications Engineering, and a doctorate in Information Engineering from the University of Padova, Italy. During his doctoral studies, he was a visiting researcher at the Centre for Wireless Communication at the University of Oulu, Finland, and at the California Institute for Telecommunications and Information Technology at the University of California, San Diego. Prior to joining the Translational Institute, he was a postdoctoral researcher at the Qualcomm Institute, University of California, San Diego. He currently serves as a reviewer for several IEEE and ACM journals, and he was the co-chair for the CQRM symposium at IEEE GLOBECOM 2015.

Contact Information

Digital Technology Center

499 Walter Library, 117 Pleasant Street SE, Minneapolis, MN 55455

P: (612) 624-9510

Connect on Social Media