|
|
楼主 |
发表于 2006-12-26 14:12:10| 字数 4,799| - 中国–上海–上海–浦东新区 电信
|
显示全部楼层
Loose Wire: New Service Identifies Songs You Hum
Jay Bose has the best customers in the world. If his service works, they're happy. If it doesn't, they blame themselves, don't ask for their money back, and try again. That's because the American chief operating officer and general manager of Nayio Media Inc., and his South Korean colleagues, have realized one thing: Most of us think we're lousy singers.
California-based Nayio has been selling a music recognition service in South Korea for a year now, allowing subscribers of SK Telecom Co. to hum a tune into their cellphone, and, for two cents, receive in return an SMS text message listing the songs it might possibly be. Last week the company launched a version for the rest of the world, available free at hsearch.nayio.com. What Mr. Bose noticed from recordings of early testers in Korea and his own experiences at a recent trade show: Unlike with speech recognition, which he had worked in before, customers didn't blame the system if it didn't understand them. 'If it got it right, they were happy; if it didn't, they just laughed about it,' he says.
The idea of a computer trying to figure out what you're singing isn't new. Cornell University developed some software in 1995 that identified the rises and falls in a tune -- its so-called melodic contour -- and compared it with a database of 183 songs. Other academics have followed in Cornell's wake, although Nayio appears to be the first commercial attempt at a service that searches for what people hum.
A similar technology, 'audio fingerprinting,' which compares slices of audio until a match is found, already is common. MusicBrainz, for example, can go through your MP3 files and try to guess which songs they are so you don't have to label them yourself. Shazam lets you hold a cellphone up to a speaker and tells you, via SMS, what song is playing. Some Sony Ericsson phones have a similar feature.
Nayio's system is a bit different. Mr. Bose says it works by looking at the melody and pitch of the song, not the actual notes. 'Instead of looking at the notes, we look at the transitions between each note,' he says. From that, the software builds a so-called genetic footprint, or code, of the song, which the company calls muGene (short for Music Gene). Each song in the Nayio database -- currently in the thousands but growing fast -- is broken down into small, 500-byte files. A customer hums 10 to 15 seconds of the song they're trying to match, and that is converted to a blob of code that can then be compared with those in the database. The songs that match most closely are listed in order, along with an estimate of the likelihood of each one being the song that was hummed, given as a percentage. It doesn't matter, in other words, whether you start on the right note, or sing in tune, or even at the right speed, so long as the notes are more or less correct relative to each other. And because everything is converted to byte-size chunks of code, the searching process is as fast as Google: a couple of seconds.
Does it work? Well, yes and no. My efforts started off well -- the Beatles''Yesterday' was identified with 77% accuracy -- but perhaps my voice was worn out by that initial 12-second burst, because all subsequent efforts were misses. Minnie Ripperton's 'Lovin' You' sounded to Nayio like Guns N Roses''You Ain't the First,' 'White Christmas' sounded like 'Amiga Mia,' and the software was 89% certain my rendition of Abba's 'Dancing Queen' was in fact 'I Am Alive' by Helloween. Even 'Jingle Bells' was misidentified as an Alanis Morrissette number. My wife had the same experience: Her first, a Corrs track, was spot on, but everything else was a palpable miss. Importantly, neither of us blamed the software; we blamed each other.
Perhaps with this lack of musical confidence in mind, Nayio also has launched another service, using the same technology. In what Mr. Bose calls 'karaoke on steroids' but I would call 'karaoke for the paranoid,' users can sing along to a song on their computer and then have the software offer comments on their performance, ranging from 'great' to 'what the...' The users can compare, on a kind of primitive musical score, the notes they have sung with the ones they were supposed to have sung. They can then practice line by line until the software stops kicking them in the figurative teeth. If you're really confident of the results, you can add a bit of video of yourself singing via a Web cam, mash it in with a bit of stock footage from Nayio's library and upload the results to the Nayio Web site for others to laugh at. If you really want to.
Given my previous experience with 'Jingle Bells,' one of the free songs offered by the service, I knew I was in for trouble, and indeed, the results were not pleasant. My wife left the room after the first verse, without clarifying whether she was ever coming back. The software was similarly harsh in its verdict, consistently rating me as either 'poor' or 'what the...'
The training idea is a nice touch, especially if you're not as confident of your vocal cords as you would like to be. The downside: Only 20 or so free songs are on offer; users in the U.S. can buy other songs from Nayio partner Napster, a service that the company plans to make available in other countries soon.
Being able to search music just by humming it could be as useful as being able to search video and images, letting us comb not just through the music on CDs and MP3 files but movies, television shows and video clips. Allowing you and me to find the song we're looking for simply by humming a snatch of the tune, however awfully, would be as big a step for the world of music as Google was for the world of text. Nayio isn't quite there yet, but it's not far off. That it may actually help us sing better is just a bonus. |
|
[复制链接]
|自动提醒
狗仔卡
离线
提升卡
置顶卡
变色卡
