MINING THE TREASURES OF PIANO ROLLS.

Byline: Edward Rothstein The New York Times

The piano is, to be blunt, a ridiculous musical instrument. It depends almost completely on sleight of hand. Each note begins with a percussive strike and immediately begins to decay.

But by distracting the ear with harmonies that push and pull, by using swells in the supporting left hand, by striking a second note milliseconds before the first ends, by invoking resonances of partially damped strings, a pianist can turn crude impacts of hammers into music.

A master of illusion can make it seem that a sustained note is gradually getting louder, that one tone is sliding liquidly into the next, that there really are vocal cords, or at least violin strings, inside that massive machine of wood and metal and felt.

But try using another machine to reproduce such magical effects. Nearly all attempts to recapture the musical ideas punched out in old piano rolls have been disastrous. The once dominant reproducing piano, with its vacuum pump and levers, has been reduced to a curiosity, even when gussied up in modern restorations.

The player mechanism betrays the mechanism of the instrument; the pianist's sleight of hand is undone. It is as if a robotic magician were clumsily pulling cards from its sleeve, or jerking a table with a fake bottom, inadvertently springing a caged rabbit before its time.

In my experience, the only credible musical reproductions have come from the relatively new Yamaha Disklavier, in which the action responds to instructions from a computer floppy disk.

Until recently, at least. Harold C. Schonberg, a former chief music critic of The New York Times, has been leading listeners through a survey of reproducing-piano performances in a radio series titled ``Earwitness'' on the nationally syndicated WFMT Fine Arts Network.

The programs, which are broadcast in New York on Monday evenings at 10 through January on WNYE, include shows devoted to Vladimir Horowitz, to women of the piano, to Stravinsky's pianism and to operatic transcriptions.

Grieg plays on a roll made before 1907; Camille Saint-Saens at the ripe age of 83, ripples through an improvisation on his ``Samson and Delilah''; the 15-year-old Shura Cherkassky toys with Liszt's ``Rigoletto.'' There are also unusual performances by Guiomar Novaes, Vladimir de Pachmann and Alexander Siloti.

Often, of course, the sonic artifice of reproducing pianos is evident: overstressed musical punctuation, narrowed ranges of timbre and dynamics, terraced and aggressive phrasing.

In these Duo-Art rolls, the infinitely refined gestures of a great instrumentalist are broken into discrete bits of information: the earliest form of digital audio. So Maurice Ravel's roll of his ``Gibet'' is cluttered with accents and poorly balanced; and although Leopold Godowsky's performance of Chopin's G minor Ballade (Op. 23) ends with a mad sweep of folkish dance, it begins with staggered, mechanistic utterances.

Still, there is something uncanny in these rolls. Schonberg, who is an expert in the early recorded history of the piano, once dismissed rolls as ``untrustworthy'' evidence of a pianist's achievement. Now he is prepared to recant. ``I have heard things I never heard before,'' he said recently. These renditions he calls ``the best that have been done in the history of recording.''

Novaes is positively luscious in her elegant, poised readings of Saint-Saens and Moszkowski. Josef Hofmann's rigorous presence is often unmistakable. Ignace Paderewski - who, as Schonberg points out, is miserably represented on early acoustic recordings - is an almost magnetic force in the Liszt-Wagner ``Liebestod.'' And de Pachmann turns Chopin's E minor Etude (Op. 25, No. 5) into a swerving study in arpeggiation spiced by rubato.