Fusion method improves speaker recognition by handling uncertainty
Entropy-aware logistic regression for fusion of large-scale speaker recognition systems
Sound
Summary
Speaker recognition systems often combine multiple sources of information to better identify who is speaking. The usual approach uses fixed settings to mix these sources, but it doesn’t consider how clear or reliable each recording is. The authors introduce a way to include information about how uncertain the system is about each recording when combining results. This new method helps make more accurate identifications, especially when audio quality or speaking conditions vary a lot. They show this uncertainty information makes a useful addition for large-scale speaker recognition.
What this means in practice
- •For voice biometric system developers: Improve accuracy in large-scale voice identification by accounting for variability in input speech quality during score fusion.
- •For call center automation teams: Enhance speaker verification robustness in customer service calls by integrating uncertainty measures into recognition system fusion.
Authors
Pierre-Michel Bousquet, Mickael Rouvier
Abstract
Score-level fusion based on logistic regression is widely used in speaker recognition to combine complementary systems. However, conventional approaches assign fixed system-dependent coefficients and do not explicitly account for variations in the reliability of individual enrollment and test utterances. Drawing on recent research on the entropy of deep learning-based speaker recognition models, this study incorporates an uncertainty component into the fusion process. By exploiting both system-level complementarity and utterance-dependent uncertainty, the method achieves robust performance in large-scale speaker recognition tasks that involve highly variable characteristics of the speech signal. These results demonstrate that model-entropy information provides a valuable complementary cue in large-scale scenarios.