Abstract :
This work identifies relevant songs from a user´s personal music collection to accompany pictures of an event. The event´s pictures are analyzed to extract aggregated semantic concepts in a variety of dimensions, including scene type, geospatial information, and event type, along with user-provided keywords. These semantic concepts are then used to form a search query against a song database based primarily on the song lyrics. Songs are scored using probabilistic techniques to come up with a rank ordered list of candidate songs that could then be used as, e.g., the audio track in a multimedia slide show.
Keywords :
audio databases; image matching; music; probability; query processing; vocabulary; event picture; geospatial information; image collection; probability; rank ordered list; scene type; search query; semantic concept extraction; song database; song lyrics; song matching; user personal music collection; Data mining; Digital communication; Image databases; Indexing; Information analysis; Layout; Multimedia databases; Pediatrics; Social network services; Video sharing; Indri; events; lyrics; multimedia; music selection; music slideshow;
