RSS

An ecological approach to the classification of transient underwater acoustic events: Perceptual experiments and auditory models

Investigator: Simon Tucker Supervisor: Guy J Brown

A research project jointly funded by QinetiQ and and the EPSRC.

Thesis abstract:

This thesis investigates the classification of underwater acoustic transients under two guiding principles. Firstly, auditory models were used to analyse the underwater signals and, secondly, an ecological approach was taken in the search for acoustic features of transient events. Three features were identified: one which assessed the timbre of the transient event, one which evaluated the material of the transient source and one which measured the temporal structure of a sequence of transient events.

The timbre of sonar transients was investigated through a multidimensional scaling study, which was used to construct a three-dimensional timbre space on the basis of listener judgements of acoustic similarity. This space was used to construct an auditory model which recreated the space derived from listener judgements and, furthermore, was used to assess the timbre of novel sonar transients. A second experiment investigated human perception of physical properties of objects when vibrating both in-air and underwater. The experimental results were then used to construct a model of material perception; this model was shown to make similar confusions to human listeners. The temporal structure of a sequence of transients was investigated using a multi-scale rhythmical representation. Specifically the distribution of inter-onset intervals at a number of scales was estimated, and used to construct a means of classifying the temporal structure of transient sequences.

The transient measures were integrated to form a single feature vector, and the performance of this vector was compared to a simple spectral representation and a set of statistical spectral measures in a relatively large classification task. The results of the study show that the novel features developed here gave improved classification performance compared to the other two feature vectors examined.