Sing For Your Search

Ever tried to use an Internet Search Engine to track down a song that you half-remember hearing once at a great halloween fancy dress party about five or six years ago? Chances are you didn’t meet with much success – typing a description of how the song sounded is likely to lead you down some very dark Web alleys indeed.

If you remember a few fragments of lyrics, you might have a far greater chance of finding that elusive tune – you could, for example, use Yahoo!’s lyrics search on their music site (as powered by a recent partnership with Gracenote). Strangely enough, a standard search in Yahoo! for ‘yahoo gracenote lyrics’ doesn’t provide any links to this service – I actually found the site through Thomas Hawk’s photography blog, which is full of useful tips for imagesmiths.

A recent article from Australia’s ABC Net mentions research from RMIT University, which claims that search technology will soon have improved to the extent that searchers can sing fragments of songs to their browser, which will be able to identify the track and return valid results. According to computer scientist Dr Sandra Uitdenbogerd, you won’t even have to sing well – it will simply take longer to make a match if you are out of tune, or if there is a lot of background noise.

I wonder if the RMIT crew are aware of such sites as Midomi, where one can already search for songs by singing or humming a fragment. The site is still in beta, but the database is growing steadily with contributions from users helping to diversify the range of information available – any genre of music can be submitted, in any language. Users can also create profiles, rate each other’s performances, discover new music and buy songs for download.

The technology behind Midomi has been developed by Melodis Corporation, who call it Multimodal Recognition System Search (MARS). This voice-activated search technology uses a variety of music-specific variables to poll their database, including pitch, tempo, pauses, modulation and rhythm.