It was the idea that almost got away, but for a coincidence. In 2002, James Hom was working on a lab assignment for Introduction to Electronics when he mentioned to teaching assistant Keyvan Mohajer that he had an idea for a search engine that could find music you hummed. Hom’s concept even won a finalist slot in the Stanford E-Challenge business competition, but nothing further came of it then.
In 2004, Mohajer, MS ’02, PhD ’07, remembered the idea, and believed the technology had evolved to the point where it could be achieved. He searched for Hom, ’05, in Stanford’s online directory and even composed an email to him, but talked himself out of sending it. “I decided maybe I shouldn’t act based on excitement,” Mohajer remembers.
The next day, Mohajer was driving on campus. At the stop sign at Campus Drive and Mayfield, he saw Hom walking in the crosswalk in front of him. Mohajer rolled down the window and asked Hom to hop in. (“He was a little taken aback,” Mohajer says.) But after a talk over coffee at the CoHo, Hom was on board.
Five years later, Mohajer is the president and CEO and Hom is the vice president for products and CFO of Melodis, a sound-search company in San Jose. Their flagship product, Midomi, is the cure for that age-old dilemma: remembering the name of that song you have stuck in your head. In fact, if you know just a few bars of a song by virtually any artist—from Rihanna to the Rolling Stones—Midomi can not only tell you what it is, but also link you to iTunes to buy the track, show you the video, and help you meet fellow fans. The key is its unique melody-based algorithm, which also incorporates tempo and lyrics but isn’t sidetracked by them.
Using Midomi is simple: just hum, whistle or sing into your PC microphone or iPhone for about 10 seconds. Midomi will search a database of nearly a million audio clips recorded by users around the world in 33 different languages and tell you the song. Wanna be a rock star? Click “studio” and record your own rendition for inclusion in the database.
Can’t carry a tune? Doesn’t matter. Mohajer hums a melody into his phone. This reporter doesn’t recognize it as Norah Jones’s “Come Away with Me,” but Midomi pops up with it right away. That’s good news, because one popular use of the search engine is in college drinking games. (Midomi doesn’t get your song? Drink! Midomi does? The other guys drink!) Under such circumstances, the search queries may sound a bit, shall we say, imperfect.
Midomi was born at the end of 2004, when Mohajer, Hom, Majid Emami and Mike Grabowski, who has since left Melodis, packed 20 computers into Mohajer’s apartment in Studio 6 at Escondido Village. Hom and Grabowski, ’04, MS ’06, went home for the December holidays, but Mohajer and Emami, MS ’02, PhD ’07, spent two weeks straight in the apartment—leaving only on Christmas Eve to eat burritos at Taqueria El Grullense on El Camino Real. When their colleagues returned, Mohajer and Hom had a prototype of their music search engine.
The following summer, Mohajer proposed having users build the music database by recording themselves singing song clips. “He said, ‘I have this idea. I am not sure if it’s a good idea, but I am going to say it anyway,’” remembers Emami, now Melodis’s vice president for engineering. It turned out to be better than good.
Midomi users not only have built a music database, but also spread the word about the service and created a community on the site. Some of them become “Midomi stars.” For example, “Fallingsta,” of Las Palmas de Gran Canaria, Spain, has attracted 312 “friends” and 209 “fans” with his 62 recordings—everything from Elvis Presley to Dionne Warwick to Coldplay, backed by karaoke-style synthesized accompaniment.
“It’s like Facebook for singers,” says Yule Schmidt, ’09, who was paid to help seed the database in 2006 and has recorded almost 350 song clips, a cappella, for Midomi. “You get comments from 11-year-old girls and people in Italy.”
At present, Midomi is free to users and supported by advertising. Melodis, which is financed by venture capital, is experimenting with the best revenue model. They may eventually charge for use, or develop a premium product for pay.
Midomi’s chief competitor is London-based Shazam, which has been in business a couple more years and boasts 8 million searchable tracks. But Shazam recognizes just the originally recorded piece—if you’re in a store, say, and want to know what’s playing, you aim your mobile phone at the stereo speakers. Midomi’s multimodal approach also lets you hum or speak or sing off key and off tempo. “If you sing it really wrong, you won’t get it, though,” Mohajer says.
Larry Marcus, a board member and investor, says when he first heard about Melodis it reminded him of 2001: A Space Odyssey, where people talk seamlessly to computers. “I started connecting the dots around what the potential might be,” Marcus says. “What I saw in this company was this incredible founding team that has developed a really superior way of using sound to search for very timely information.”
Under that view, Midomi is just the beginning. Another Melodis product, Dialer, lets you dial your iPhone using voice instructions, rather than scrolling through a long list of contacts. Traditional voice dialing technologies convert speech to text. Dialer accepts the sound directly, making the search faster and more accurate. Future Melodis products might monitor heart rate or allow you to search the Internet while driving.
What’s the future of a company like this—whose main product isn’t strictly necessary—in lean times? Marcus says it’s still bright: “Even in hard economic times, people want to find joy. One of the real sources of passion and joy in people’s lives is music.”
This article was modified from the print version of the magazine.
CHRISTINE FOSTER, of Saratoga, Calif., is a Stanford contributing writer.
