Spectra transformed for model-testing and visual exploration

Abstract

The presence of highly tangled patterns in spectra and other serial data exacerbates the difficulty of performing visual comparison between a test model for a particular pattern and the data. The use of a simple map that plants peaks in the data directly onto their corresponding position in a residual plot with respect to a chosen test model not only retrieves the advantages of dynamic regression plotting, but in practical cases also causes patterns in the data to congregate in meaningful ways with respect to more than one reference curve in the plane. The technique is demonstrated on a polyphonic music signal.