On Calibration of Speech Classification Models: Insights from Energy-Based Model Investigations

TL;DR

This paper investigates the use of Energy-Based Models to improve the calibration of speech classification models, addressing overconfidence issues and enhancing decision reliability across multiple speech tasks.

Contribution

It introduces a joint EBM approach that combines discriminative and generative models to improve calibration in speech classification without losing accuracy.

Findings

EBMs effectively improve calibration in speech models

The approach reduces overconfidence in classifiers

Calibration enhancement achieved across multiple speech tasks

Abstract

For speech classification tasks, deep learning models often achieve high accuracy but exhibit shortcomings in calibration, manifesting as classifiers exhibiting overconfidence. The significance of calibration lies in its critical role in guaranteeing the reliability of decision-making within deep learning systems. This study explores the effectiveness of Energy-Based Models in calibrating confidence for speech classification tasks by training a joint EBM integrating a discriminative and a generative model, thereby enhancing the classifiers calibration and mitigating overconfidence. Experimental evaluations conducted on three speech classification tasks specifically: age, emotion, and language recognition. Our findings highlight the competitive performance of EBMs in calibrating the speech classification models. This research emphasizes the potential of EBMs in speech classification tasks, demonstrating their ability to enhance calibration without sacrificing accuracy.