Kernel Regression Mapping for Vocal Eeg Sonification

This paper introduces kernel regression mapping sonification (KRMS) for optimized mappings between data features and the parameter space of Parameter Mapping Sonification. Kernel regression allows to map data spaces to highdimensional parameter spaces such that specific locations in data space with pre-determined extent are represented by selected acoustic parameter vectors. Thereby, specifically chosen correlated settings of parameters may be selected to create perceptual fingerprints, such as a particular timbre or vowel. With KRMS, the perceptual fingerprints become clearly audible and separable. Furthermore, kernel regression defines meaningful interpolations for any point in between. We present and discuss the basic approach exemplified by our previously introduced vocal EEG sonification, report new sonifications and generalize the approach towards automatic parameter mapping generators using unsupervised learning approaches.