Music Informatics

Members

Research focus topics and daily updates can be found on members' personal websites.

Research Network

Activities

Development of AMUSE (Advanced MUSic Explorer)

  • Java open source framework for different MIR tasks
  • Multi-tool extendable environment with shared interfaces and data interchange formats
  • Algorithm evaluation and optimization methods

CI Methods for Music Information Retrieval

  • Optimization of music classification chain
  • Intelligent feature selection methods
  • Learning of personal music categories
  • Fuzzy methods using high-level feature analysis

Music Test Database

For the evaluation of music classification tasks we use a database of 120 commercial music albums purchased for our research group. The CDs are distributed among 6 AllMusicGuide genres:

Classic Electronica Jazz Pop/Rock R&B Rap
CD number 15 15 15 45 15 15

  • Test set OS120 (optimization set for experiments, with focus on evolutionary multi-objective feature selection)
  • Test set TS120 (song-independent test set from the same albums as OS120 for the evaluation of optimized models)
  • Test set TAS120 (artist-independent test set with the same genre distribution)

The following audio features are used for classification (for definitions and the difference between low-level and high-level features see https://eldorado.tu-dortmund.de/handle/2003/30402)

Instrument Sample Database

Currently, the following instrument samples are used for the experiments on instrument recognition:

 
Last modified: 2023-08-05 17:47 by Fabian Ostermann
DokuWikiRSS-Feed