da/sec scientific talk on Biometrics

Topic: Speaker Recognition in unconstrained Environments

by Andreas Nautsch
D19/2.03a, November 3, 2016 (Thursday), 12.00 noon

Keywords — voice biometrics, machine learning, domain adaptation, quality-based calibration

Abstract

By increasing demands for biometric systems in mobile application scenarios, higher demands are posed to biometric systems, such as speaker recognition. In the presentation, recent research advances regarding unconstrained environments will be put to discussion. Environmental conditions may vary depending on e.g., speech quality, spoken texts and languages, and background noises. Our approach to cope with changing data domains utilizes quality estimates employed in score calibration for improving an overall decision cost.