Supplementary Material for "Listen to your Data: Model-Based-Sonification for Data-Analysis"

Sonification is the use of non-speech audio to convey information. We are developing tools for interactive data exploration, which make use of sonification for data presentation. In this paper, model-based sonification is presented as a concept to design auditory displays. Two designs are described: (1) particle trajectories in a data potential is a sonification model to reveal information about the clustering of vectorial data and (2)data-sonograms is a sonification for data from a classification problem to reveal information about the mixing of distinct classes. Sound Examples 2+3 with Model II Data-Sonograms: These examples present Data-Sonograms of the iris data set. Iris_nm is a sonogram where the shock wave starts at the mean of all instances of class n. The local entropy S is computed using instances of class n and m. Thus the class border between class n and m is sonified.

声化（Sonification）指利用非语音音频传递信息。本团队正研发面向交互式数据探索的工具，此类工具借助声化技术实现数据展示。本文提出基于模型的声化概念，用于设计听觉显示方案。文中详述两种设计方案：其一为数据势场中的粒子轨迹声化模型，用以揭示矢量数据的聚类特征；其二为数据声图（Data-Sonograms）声化方法，针对分类任务中的数据，用于展现不同类别间的混合特征。

搭配模型II数据声图的示例音频2与3：此类示例展示了鸢尾花数据集（Iris Dataset）的数据声图。其中Iris_nm为一类声图，其冲击波起始于类别n所有样本的均值。局部熵S通过类别n与m的样本计算得到，由此实现类别n与m之间的类别边界的声化呈现。