 |
 |
 |
    |
 |
 |
 |
What's new. What's hot. What they're saying. |
 |
|
 |
 |
 |
|
|
|
 |
|
||
|
TECH: Will Elvis Return, Chopped to Bits? The job of singer may soon be outsourced. A new "vocaloid" is being called the "world's first virtual soul vocalist," and he has lots of friends on the way. Meanwhile, other scientists are trying to teach computers to hear the different between speech and song.
Here comes Leon. And Lola. And Miriam. "They" are the first commercial realizations of Yamaha's "vocaloid" technology, which it has licensed to UK's Zero G and other companies. For around $300, you now can own one of these virtual singers, who can be instructed by your PC to sing a set of notes and words. A recent article in the New York Times characterized these vocaloids as "audio fonts": assemble lyrics, assign them musical notes (pitch, timing, expression, etc.), and then choose a voice to sing them. Comparing this new technology to earlier attempts to capture the subtlety of human expression, the Times concludes that "the company seems to have made that quantum leap." A Japanese company will be releasing its first audio font in March, a bubbly female Japanese pop voice. So far, no famous singers have indicated that they will volunteer to be transformed into vocaloids. (E-mail us your thoughts about who will be the first!) Meet Leon & Lola at Zero-G's site, and listen to demos of their singing Read Yamaha's March 2003 press release, or visit the company's product page
What about the other direction? That is, how well do computers hear singing? So far, not well. Until recently, computers could not tell the difference between a spoken or sung version of the same words, a task that even a baby can do with no effort at all. That's changing, thanks to the work of musician & computer scientist David Gerhard, a professor at Canada's University of Regina. His work has uncovered measurable differences between singing and speech, which is what computers need in order to "compute" what they're listening to. His technology may be used to help computers understand and react to their users needs, based on what users are doing. Learn more about Gerhard's work (including an audio demo) and the list of songs his computers have been struggling with Browse Gerhard's 2003 PhD thesis [pdf format] from Simon Fraser Univ.: "Computationally Measurable Temporal Differences between Speech and Song" |
|
