April 20, 2015 | Gisèle Bolduc
Update : November 25, 2020
Professors Tiago Falk and Douglas O’Shaughnessy of the INRS Énergie Matériaux Télécommunications Research Centre have each received a Google research grant to pursue their innovative work in the field of speech processing. These grants will allow two PhD candidates in telecommunications to take part in research for a year and give them the opportunity to collaborate with Google researchers and engineers.
How can we make automatic speech recognition (ASR) more accurate and reliable? How can we reduce the detrimental effects of background noise, room reverberation, and competing speakers? How can we objectively quantify the gains of such speech enhancement strategies? Students from the teams of professors Falk and O’Shaughnessy will explore new ways of improving and measuring speech quality for various voice command applications, including hands-free telephony and smart TV.
Professor Falk’s team will perfect the in-house developed ‘speech-to-reverberation modulation energy ratio’ (SRMR) non-intrusive quality measure so it’s applicable to burgeoning adaptive microphone arrays. The goal is to create an instrument that can quickly, reliably, and accurately quantify the perceptual effects of microphone array based speech enhancement algorithms and update internal algorithmic parameters in real-time to maximize the perceived quality.
Professor O’Shaughnessy’s team will work on developing a new method to analyze acoustic speech by focusing on the temporal structure of spectral resonance. His approach will be based on obtaining a better understanding of speech production and perception in order to simplify analysis.
These two projects were chosen from among over 800 applications submitted during Google’s last semi-annual call for proposals. Only five projects by Canadian researchers were selected.
The Google research awards program was created to support the work of world-class faculty members in fields such as computer science, robotics, and human-computer interaction in computational neuroscience. These one-year grants can cover tuition for a student for one year.