Computer aided design of audio signal classifier systems

Abstract

The problem of programming classifier systems for deterministic events in bioacoustic signals is seen as a knowledge engineering problem. Advantages and disadvantages of traditional approaches are discussed. A knowledge engineering approach for interactive visual classifier system design is described and advantages of the approach are highlighted. The basic architecture of a general purpose knowledge engineering framework for the acoustics domain is outlined. Emphasis is placed on both expert knowledge and on the role of data mining within the classifier system design process. Several examples illustrate how the approach can be used to create classifier systems for bioacoustic patterns in the time‐frequency domain. Examples include multiscale spectrographic visualization, techniques to create and interactively modify subsymbolic and symbolic classifier systems for time‐frequency patterns as well as the usage of three‐dimensional visualizations to display hierarchical dependencies within complex classifier...