Abstract by Xin Zhao
Physics and Astronomy
An automated pipeline for machine learning in crowd acoustics
Acoustic signals can provide insights into the state of a crowd. For example, a crowd at a sporting event could be engaged, happy, upset, or indifferent. Their level of engagement is reflected in the acoustic sounds they make. The acoustic signal may also be connected to the specific events in the game. We are using machine learning to connect the acoustic signal coming from a crowd with their emotional state and the events that happen in sporting events. Recordings of the crowd have been collected at BYU basketball, volleyball, soccer, and football games. The scope and diversity of this data requires sophisticated data management strategies. I describe efforts to develop a data-pipeline management system for machine learning in crowd acoustics, so that we can accelerate the entire process of machine learning in crowd nosie. Our goal is to automate data pre-processing and make the processed data available to a team of researchers in a user-friendly environment.