Perceptually-Driven Signal Analysis for Acoustic Event Classification
Abstract
This research develops a framework for employing perceptual information from human listening experiments to improve automatic event classification. We focus on the identification of new signal attributes, or features, that are able to predict the human performance observed in formal listening experiments. Using this framework, our newly identified features have the ability to elevate automatic classification performance closer to the level of human listeners. We develop several new methods for learning a perceptual feature transform from human similarity measures. We also develop a new approach for learning a perceptual distance metric. Our research demonstrates these new methods in the area of active sonar signal processing and confirms anecdotal evidence that human operators are adept in the task of discriminating between active sonar target and clutter echoes. We identify perceptual features and distance metrics using our novel methods. The results show better agreement with human performance than previous approaches.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 26, 2007
- Accession Number
- ADA476786
Entities
People
- Jack Mclaughlin
- James Pitton
- Scott Philips
Organizations
- University of Washington