University of Bamberg
open search form
You are here

Information Systems and Applied Computer Sciences

Chair of Computational Humanities

  1.  University of bamberg
  2. Faculties
  3. Information Systems and Applied Computer Sciences
  4. Subject groups
  5. Applied Computer Science
  6. Chair of Computational Humanities
  7. Research
  8. Datasets

Datasets

Jakob Abeßer has been actively involved in the creation and publication of a large number of research datasets in the field of audio and music analysis; three of these are presented in more detail on this page as examples.

Synthetic Pitch Contour (SPC) Dataset 

The SPC dataset was created in 2024 at the International Audio Laboratories Erlangen in collaboration with Simon Schwär and Prof. Dr. Meinard Müller from the Chair of Semantic Audio Signal Processing at Friedrich-Alexander University Erlangen-Nuremberg. It contains 3,500 short audio clips with synthetically generated fundamental frequency contours from the seven contour classes: stable (stable fundamental frequency), alternating (varying fundamental frequency, such as in a siren), vibrato (periodic fundamental frequency modulation, e.g., in classical singing), glissando (continuous frequency transition), bend (one-time frequency modulation), sawtooth (frequency modulation corresponding to a sawtooth function), and triangle (frequency modulation corresponding to a triangular wave). The dataset enables the training and evaluation of AI models for classifying fundamental frequency contours and has been successfully applied to datasets from various audio domains (music, speech, everyday sounds, animal sounds). The dataset will be published in the coming months (please contact us if you are interested).

Reference: 

  • Abeßer, J., Schwär, S., & Müller, M. (2025). Pitch contour exploration across audio domains: A vision-based transfer learning approach. arXiv. https://arxiv.org/abs/2503.19161

Urban Sound Monitoring (USM) Dataset

The USM dataset was developed as a reference for various subtasks in environmental noise analysis, including source separation, sound event detection and localization, and sound polyphony estimation. The focus is on typical soundscapes in urban environments. The dataset comprises 24,000 short, synthetically generated stereo soundscapes. These were created from mixtures of 2 to 6 individual sounds that differ in volume and spatial positioning within the stereo field.

Further information:

Reference

Weimar Jazz Database (WJD)

The Weimar Jazz Database (WJD) contains 456 transcriptions of improvised jazz solos. The focus is on single-part wind instruments such as the saxophone (alto, tenor, soprano, baritone), trumpet, trombone, and clarinet. The selection of musicians and pieces reflects the evolution of jazz history—from traditional jazz and swing through bebop, cool jazz, and West Coast jazz to hard bop, modal jazz, and post-bop. For each solo, there are handwritten annotations of the main melody, playing techniques, beat positions, and harmonic changes, as well as extensive additional metadata about the transcribed recording. For copyright reasons, the original recordings cannot be made available directly via the Jazzomat Research Project website. However, the corresponding recordings can be accessed via YouTube links through the online application JazzTube (developed by Stefan Balke and Meinard Müller).

Further information:

Reference:

  • Pfleiderer, M., Frieler, K., Abeßer, J., Zaddach, W.-G., & Burkhart, B. (Eds.). (2017). Inside the Jazzomat – New perspectives for jazz research. Schott Campus. https://schott-campus.com/jazzomat/

Additional Datasets