Unleashing the Potential of Data Analytics Through Music

Abstract

One of the most significant challenges with sequential data is the identification of certain underlying patterns, which can be easily overlooked during human visualization. To overcome such problem, transforming data into sound (Sonification) has shown great potential to notify human about hidden patterns. Sonification is a process of mapping data into a non-speech audio format. In today’s era, with data analysis being the backbone of most infrastructures, Sonification is gaining interests in various fields like data mining, human-computer interaction, exploratory data analysis, and musical interfaces. It presents a novel way of analyzing and interacting with data. In addition, it provides visually impaired people with an attainable alternative. A considerable amount of work has been done in the field of sonification and music generation. However, producing music from data using machine learning and deep learning techniques is inadequately explored. The conventional sonification methods require human involvement and knowledge of music to produce a tune that is appealing to the ears. This method is time-consuming and requires specialized expertise. In this paper, we aim towards developing a system that molds any time-dependent data into music while retaining the original characteristics of the data using deep learning techniques such as LSTM (Long-short Term Memory). The goals of this research are 1) generate music that is melodious and resembles the music composed by humans, and 2) help people not only auralize but also understand the associated data trend through the generated music. Quantitative and qualitative evaluations were used to validate our approach.