Date of Completion

5-10-2014

Embargo Period

5-9-2014

Advisors

Dr. Heather Read, Dr. Krystyna Gielo-Perczak

Field of Study

Biomedical Engineering

Degree

Master of Science

Open Access

Open Access

Abstract

Environmental sounds, both man-made and natural, vary on multiple time and frequency scales generating a large range of temporal, spectral and amplitude modulations that are evident in the high-order statistics of the sound spectrogram. Healthy hearing humans perceive high-order statistical regularities and use this information to categorize and discriminate sounds. This paper tests the hypothesis that biologically motivated sound statistics can enable/enhance discrimination and identification of sound categories from a computational standpoint. A large catalogue of natural and man-made sounds and their associated high-order contrast and intensity statistics were developed, and the information carrying content of each statistic for sound recognition and discrimination was measured. Bayesian classification and signal detection theory were applied to the sound database to identify statistics that can be used to categorize sounds and to test discrimination limits amongst sounds or categories. The catalogue will be deployed as an online database available to researchers and scientists.

Major Advisor

Dr. Monty Escabi

Download

Share

COinS