Self-paced ensemble learning for speech and audio classification

TL;DR

This paper introduces a self-paced ensemble learning method for speech and audio classification, where models iteratively learn from each other, leading to improved performance across multiple audio tasks.

Contribution

The paper proposes a novel self-paced ensemble learning scheme that enables models to learn collaboratively through iterative pseudo-labeling, enhancing overall accuracy.

Findings

SPEL significantly outperforms baseline ensemble models.

Self-paced learning on individual models is less effective than ensemble learning.

Models in the ensemble learn from each other, improving domain adaptation.

Abstract

Combining multiple machine learning models into an ensemble is known to provide superior performance levels compared to the individual components forming the ensemble. This is because models can complement each other in taking better decisions. Instead of just combining the models, we propose a self-paced ensemble learning scheme in which models learn from each other over several iterations. During the self-paced learning process based on pseudo-labeling, in addition to improving the individual models, our ensemble also gains knowledge about the target domain. To demonstrate the generality of our self-paced ensemble learning (SPEL) scheme, we conduct experiments on three audio tasks. Our empirical results indicate that SPEL significantly outperforms the baseline ensemble models. We also show that applying self-paced learning on individual models is less effective, illustrating the idea that models in the ensemble actually learn from each other.