About this document

Active Learning for Speech Recognition by Ognjen Kundačina is a document available to read on EtoBox.

This paper presents a two-stage active learning pipeline for automatic speech recognition (ASR) that combines unsupervised and supervised methods. The first stage utilizes x-vectors clustering to select diverse samples from unlabeled speech data, while the second stage employs a batch active learning method with Bayesian inference to identify the most informative samples for labeling. The proposed approach aims to optimize data utilization and reduce labeling efforts, demonstrating superior performance comp

Author

Ognjen Kundačina

Language

EN