Interactive Media Systems, TU Vienna

High-dimensional Feature Selection for Audio Datasets

Master Thesis, Bachelor Thesis, or Practicum

About This Topic

Feature selection is the task of selecting features that best represent a given dataset. In media data (audio/video) features are usually high-dimensional representing some meaningful information as a group. Following, it is desired to keep the features as a whole rather than to tear it apart by single components. Various statistical and information theoretic approaches for grouped feature selection exist.

Additional Information

In this work state-of-the-art methods for grouped feature selection in the audio domain should be compared. Novel methods can be implemented and tested as desired. Platform: MATLAB