Music Information Retrieval for the FM4 Soundpark

What is Music Information Retrieval? The rapidly growing amount of music available in digital form via internet or digital libraries calls for entirely new computer-based methods for analysing, describing, distributing, and presenting music. The currently emerging research and application field known as Music Information Retrieval (MIR) is a direct response to that need. Over the past years, our research group at the Austrian Research Institute for Artificial Intelligence (OFAI) has accumulated substantial expertise in intelligent music processing. What is the FM4 Soundpark?

The FM4 Soundpark is an internet platform of the Austrian public radio station FM4. Artists can upload and present their music free of any charge. Visitors of the website can listen to and download all the music at no cost. The FM4 Soundpark attracts a large and lively community interested in up and coming music. Results

The goal of this project is to provide visitors of the FM4 Soundpark with new and interesting ways to interact with the Soundpark data base. Our results so far are:

  • Music Recommendation: The core of all applications is a content based music similarity function. The similarity is automatically computed and based on models of the songs' audio content. Musical instruments and voices exhibit specific frequency patterns in the audio signal. These patterns are estimated with statistical models and used to compute the audio similarity.
  • Soundpark Player: Whenever a visitor of the Soundpark listens to a song, a recommendation of five or more similar songs is provided. These recommendations are visualized in a graph-based representation. Users can interactively explore the similarity space by clicking on songs in the recommendation graph.
  • Soundpark 3D: The entire database of songs in the Soundpark is visualized as an audio landscape of sea and islands. Songs from the same genre inhabit the same islands, within islands songs are grouped according to their audio similarity. Users can travel through the landscape and explore their own audio path through the data base.
  • Automatic generation of playlists: Visitors of the Soundpark can choose a start and an end song from the data base and a playlist of eight more songs "in-between" is automatically computed. The playlist is a smooth transition from the start to the end song. This functionality is not online any more.

  • Schnitzer D., Flexer A., Widmer G.: A Filter-and-Refine Indexing Method for Fast Similarity Search in Millions of Music Tracks, in Proceedings of the 10th International Society for Music Information Retrieval Conference (ISMIR'09), Kobe, Japan, 2009.

  • Seyerlehner K., Flexer A., Widmer G.: On the Limitations of Browsing Top-N Recommender Systems, in Proceedings of the 3rd ACM Conference on Recommender Systems, New York, USA, October 22-25, 2009.
  • Seyerlehner K., Knees P., Schnitzer D., Widmer G.: Browsing Music Recommendation Networks, in Proceedings of the 10th International Society for Music Information Retrieval Conference (ISMIR'09), Kobe, Japan, 2009.
  • Flexer A., Schnitzer D.: Album and Artist Effects for Audio Similarity at the Scale of the Web, Proceedings of the 6th Sound and Music Computing Conference (SMC'09), Porto, Portugal, 2009.
  • Gasser M., Flexer A.: FM4 Soundpark: Audio-based Music Recommendation in Everyday Use, Proceedings of the 6th Sound and Music Computing Conference (SMC'09), Porto, Portugal, 2009.
  • Flexer A., Schnitzer D., Gasser M., Widmer G.: Playlist Generation using Start and End Songs: Proceedings of the 9th International Conference on Music Information Retrieval (ISMIR'08), Philiadelphia, USA, 2008.
  • Gasser M., Flexer A., Widmer G.: Streamcatcher: Integrated visualization of music clips and online audio streams, Proceedings of the 9th International Conference on Music Information Retrieval (ISMIR'08), Philiadelphia, USA, 2008.

Research staff

Partners

Sponsor

Key facts