Harmonic Sonification of High-Dimensional Continuous Data

High-dimensional continuous data is prevalent across empirical sciences but can be difficult to visualise. Sonification is a compelling alternative with many sonification approaches leveraging custom mappings of specific data features to auditory features. However, this requires 1) prior domain knowledge about the problem of interest and 2) experience with sonification for choosing appropriate auditory features. This makes it difficult to use these approaches as general-purpose methods for sonifying high-dimensional continuous data. In this paper, we propose a general-purpose sonification method that exploits the structural similarity between principal component analysis (PCA) and the amplitudes of partials in harmonic overtone spectra. In both cases, the first components are more relevant and tend to have larger values, while later components are less relevant and tend to have smaller values. Our method is easy to use and does not require any prior knowledge about the problem domain or sonification techniques. We provide a Python library and intuitive web application 1 with tutorial to further increase accessibility. In a user study (n = 11) we show that our sonification method conveys relevant information , allowing users to detect outliers with 92% accuracy solely based on the generated sound signatures.