Uncertainty quantification for acoustical problems

Abstract

Acoustical parameter estimation is a routine task in many domains and is typically done using signal processing methods. The performance of existing estimation methods is affected due to external uncertainty and yet the methods provide no measure of confidence in the outputs. Hence it is crucial to quantify uncertainty in the estimates before real-world deployment. Conformal prediction is a simple method to obtain statistically valid prediction intervals from an estimation model. In this work, conformal prediction is used for obtaining statistically valid uncertainty intervals for various acoustical parameter estimation tasks. We consider the tasks of DOA estimation and localization of one or more sources in an acoustical environment. The performance is validated on plane wave data with different sources of uncertainty including ambient noise, interference, and sensor location uncertainty, using statistical metrics. Results demonstrate that conformal prediction is a suitable and easy-to-use technique to generate statistically valid uncertainty quantification for acoustical estimation tasks.