Making Wavelets; June 1993; Scientific American Magazine; by W. Wayt Gibbs; 2 Page(s)

I am Doctor Brahms, Johannes Brahms." The voice, raspy but intelligible, comes from the speakers of Ronald R. Coifman's desktop computer. It is soon followed by an excruciating cacophony of hiss, pops, static and tones distorted to the limits of perception. "That was indeed Johannes Brahms playing his First Hungarian Dance," says Coifman, a professor of mathematics at Yale University. But the performance, recorded in 1889 on a wax cylinder donated by Thomas Edison (and since lost) and then rerecorded from a staticridden radio broadcast of a 78-RPM record made from the cylinder, is utterly unlistenable.

Or was, at least, until Coifman used a powerful and increasingly important mathematical tool known as adapted waveform analysis to strip out selectively the random noise in the recording while preserving the highly structured music beneath. With a click of his mouse, Coifman plays the cleansed version. Although not all the notes are correct in time or pitch--the original wax cylinder had melted a bit by the time it was copied--the noise is nearly eliminated. Yale musicologist Jonathan Berger and graduate student Charles Nichols are now dissecting the recording to determine whether the syncopation in it is caused by distortion or whether Brahms had a particularly jazzy style at the keyboard.