Artificial neural networks seem almost as enchanted by music as their real counterparts inside human heads. Shortly after research on neural nets kicked into high gear in the 1980s, scientists discovered that these brainlike circuits can learn music just by "listening" to various examples. The network quickly begins adjusting its internal connections, becoming better and better at handling musical patterns.
Now, professional musicians and university scientists believe that neural nets may also be creative. David L. Wessel, head of the University of California at Berkeley's music technology center, has a so-called listening assistant that spurs improvisation: It snatches a fragment of a jazz performance, elaborates on it, then injects the new structure back into the performance. After Russell C. Eberhart, a biomedical researcher at Johns Hopkins University, put his neural net on a diet of folk songs, it synthesized an original tune. And next year, Japan's NEC Corp. expects to unveil a computerized karaoke that will let nonmusicians compose their own scores: Just hum the melody, and the karaoke will add harmony.