This project area concerns the querying of digital music. Just like we can query text in a file or on the Internet, we should be able to search music.
One may distinguish among three means of retrieving music:
- Querying of the textual meta data that describes the music (album name, artist, year, etc.)
- Querying based on "use" data: based on how a piece of music has been used, it is possible to identify related music. For example, a co-play relationship between pairs of songs can be established based on playlists.
- Querying based on the music signal itself. Here, a song is typically represented as a sequence of high-dimensional feature vectors.
One may retrieve music from different sources and with different scope:
- An mp3 player
- A media server
- A web site
- The Internet
Queries of interest include range and k nearest neighbor queries that are well known in other contexts as well. But queries also include a range of novel playlist queries.
Projects in this area may involve the design of new query processing techniques, the prototyping of such techniques, and the experimental evaluation of such techniques.
We already have lots of music (both 30-second clips and full songs).
The projects will relate to an on-going research project, termed intelligent sound.
For further information, contact Christian S. Jensen.