A Non-Sequential Approach for Browsing Large Sets of Found Audio Data

I will present and demo the first stages of a novel approach for non-sequential browsing of large amounts of audio data. The aim is to make it possible to utilize found data, meaning data that was not recorded with the purpose of being used in (speech) research. Examples of found data can include archive data, radio and television speech, and interviews. We use dimensionality reduction algorithms (SOM, t-SNE) to project high-dimensional representations of audio segments onto a 2D-map. From this a simple interactive interface is constructed so a user can browse the audio efficiently. The method is still under development and there are many directions to explore, so any feedback is appreciated. As a relatively new PhD student on TMH I will also very briefly introduce myself and my background.