Senior Lecturer Birmingham City University Birmingham, England, United Kingdom
Abstract: In this paper, we propose a spectroscopic diagnostic method to generate an audio output in order to discriminate between two classes of data, based on the features of spectral datasets. To do this, we first perform spectral pre-processing and extract appropriate features from the spectra, and then apply different selection criteria to narrow down the number of features selected. To optimise the process, we compare three selection criteria, as applied to two spectroscopy food datasets in order to evaluate the performance of sonification. Lastly, the salient features are mapped to the parameters of a frequency modulation synthesizer, so as to generate audio samples. The results indicate that this approach allows user to consistently discriminate between two classes of spectral data.