Launching in beta mode on Friday, Midomi allows people to search for a song by singing, humming or whistling a bit of the tune. The site then offers search results that include commercially recorded tracks or versions of the song recorded by others who have used the site. The technology also lets people listen to the exact section of each of the results that matched their voice sample.
People also can type in a song title or artist to get results. The system recognizes misspelled words.
Melodis, the company behind the site, has licensed 2 million digital tracks that can be purchased and has accumulated about 12,000 more from users. Users, who range from aspiring American Idol contestants to professionals, can create profiles and rate one other's performances on the ad-supported site.
The underlying speech- and sound-recognition technology, dubbed Multimodal Adaptive Recognition System, or MARS, differs from similar technologies in that it looks at a variety of factors for recognizing samples, including pitch, tempo variation, speech content and location of pauses, said Chief Executive Keyvan Mohajer, who has a Ph.D. in sound- and speech-recognition from Stanford University.
Like search giant Google, Melodis was started in a dorm room in Stanford--only the idea for Melodis came a bit later: 2004.