Segmentierung von Audiodaten

The automatic structuring of audio data streams, meaning their division into sections that each form meaningful units, represents an important and crucial preprocessing step in several areas, such as automatic speech recognition or the creation of audio summaries. Good segmentation can also be highly beneficial for navigating longer audio data streams or visualizing them. This work provides an overview of existing approaches and presents three methods in detail, comparing each one that utilizes unsupervised, semi-supervised, and supervised machine learning techniques, respectively. The ability to adapt to various tasks and audio sources of the data-driven methods was of fundamental importance.