Hierarchical Clustering Used to Discover Second-Order Patterns in Speaker Recognition

This research proposes discovering second-order patterns—latent patterns underlying human-defined recognition features—in speaker recognition networks. The method uses hierarchical clustering and Hierarchical Cluster-Class Matching to semantically interpret these learned representations.