Estimating speech intelligibility with neural networks.
Bernd Accou
When?
Thursday, 22 October 2020, 14:00
Location
Seminar room HP 7
O&N 2 - KU Leuven
Herestraat 49
B-3000 Leuven
Online access
The research webinar can be accessed through the following link: https://zoom.us/j/91640521161?pwd=b1I0WElLaGFQWlNFaDA1T09RallUQT09
Meeting ID: 916 4052 1161
Passcode: 5aVXPn
Abstract
Current tests to measure whether a person can understand speech require behavioral responses from the person, which is not always possible in practice (e.g. young children). Therefore there is a need for objective measures of speech intelligibility. Recently, it has been shown that speech intelligibility can be measured by letting a person listen to natural speech, recording the electroencephalogram (EEG) and decoding the speech envelope from the EEG signal. We defined a different paradigm wherein we match/mismatch speech envelope segments to EEG with neural networks instead of decoding speech envelopes from EEG. In this research seminar, I will present the results of our new approach and our dilated convolutional model, as well as a new method to estimate speech reception thresholds.
Please reply to bernd.accou@kuleuven.be if you are planning to attend the presentation in person.
In order to comply with current Covid-19 regulations, only the supervisory committee, the presenting PhD student and 18 other people are allowed to attend the seminar in person. Thank you for your understanding.
