Printer friendly version
Cebit 2012: Intelligent software assigns appropriate background music for pictures
16 February 2012
“Usually, directors select consistent melodies for a particular movie scene,” explains Sebastian Michel, head of a junior research group at the Cluster of Excellence at the Saarland University. He designed the Picasso software together with Ph.D. student Aleksander Stupar. The program utilizes the expertise of movie producers in assigning appropriate background music for pictures, enabling users to select music that will harmonize with their impressions of their last summer vacation without putting in an extraordinary amount of effort.
A three-level algorithm carries out the elaborate process. First, the picture the user has chosen is compared with a huge database of movie scenes and their corresponding soundtracks. The software creates a ranking of the scenes that look most similar to the user’s picture. This database was compiled by the two researchers Michel and Stupar by splitting 50 movies into screenshots and their accompanying soundtracks. In the next step, the software creates a list of the selected tracks. And finally, the total number of tracks gets reduced to a few selections through a mathematical calculation; these are at last proposed to the user.
Here, a list of suggestions makes sense because particular pictures bring up different associations for different users, according to Michel. “Some people might connect a picture of a little house surrounded by an idyllic landscape with a romantic weekend for two, while others might think about loneliness,” Michel explains.
Interested people can experience the software on the demo website and get an idea of Picasso’s hit rate. Or they can use the free smartphone app “PicasSound,” which is programmed to pick out an appropriate soundtrack using the music which is already saved on the smartphone. It is free available for the mobile operating systems iOSX and Android.
Next, Michel and Stupar will try to expand the software system to support automatic sound recording of audiobooks on a textual basis. Furthermore, the researchers are planning to integrate a function inside the system that will take the individual preferences of the users into account.