Visual Charting of Classified Audio Data

Abstract

To analyze hundreds of hours of audio recorded in field testing, it is useful to use previously trained machine learning algorithms that can classify the data with discrete time-stamped events. However, this is only a first step. A programmatic method is needed to make sense of the thousands of classified events. In a field testing environment, new charts need to be generated quickly so that conclusions can be drawn while the field test is still ongoing. The generation of these charts also needs to be flexible enough to quickly respond to any on-the-fly changes. This thesis describes the development of a highly customizable method to visually chart labeled audio data in an easily understandable format.