Skip to content

Investigating Layer-Specific Performance in Speaker Recognition with XLS-R Architecture

Notifications You must be signed in to change notification settings

slinusc/speaker_recognition_evaluation

Repository files navigation

Abstract

This study explores the impact of various layers of the XLS-R encoder on speaker recognition accuracy using CNNs and logistic regression. Findings indicate that earlier layers yield higher accuracy, highlighting their importance in feature capture. The study also reveals a significant gender disparity in accuracy. These results suggest the need for further investigation into model biases and optimizations.

About

Investigating Layer-Specific Performance in Speaker Recognition with XLS-R Architecture

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published