=================
                                         The Piano project
                                         =================
Abstract:
The perceived difference in a piano performance between a Sergei Rachmaninoff and a Lang Lang versus 
a Vladimir Horowitz and a Yundi Li resides in a series of quantitatively defined sonic events. As such,
any note played can be numerically described by that note’s relative pitch/frequency and by any note’s 
timing with its beginning termed “note on” and its end point termed “note off”. Likewise, the 
perceived loudness of any note can be numerically defined by the terminal velocity of the piano 
hammer striking each string. When mapped on a grid, these values begin to reveal what in essence is 
the musical DNA of each artist in a graphical and visually striking way. In turn, if sufficient 
performance data is gathered with sufficient accuracy it is then likewise possible to minutely 
analyze and even perfectly “re-perform” any artist’s original performance. 

 Accomplishing such a “re-performance” inevitably requires the conversion of audio wav files to 
highly accurate midi (i.e. musical instrument digital interface) files which can then be played back 
on a modern computer equipped, solenoid driven piano such as those offered by Yamaha’s renowned 
Disklavier line of instruments. 

Algorithms presently exist that attempt to “extract” midi data from music wav files however the 
complex polyphonic and over-tonal nature of the piano has proven problematic with regards to accurate 
detection with the resulting midi files routinely a cluster of chaotic notes and sounds. Classical 
music, by its very nature, requires the pianist to play from a score where the composer’s intentions 
are meticulously defined. It is the slight variations in note timings and note velocities as 
determined by a particular artist’s personalized reading of the printed score that distinguish one 
performer’s rendition/interpretation from another performer’s rendition/interpretation.  
Since an accurate score based midi file can also be readily generated, the proposed project would 
attempt a probabilistic reconciliation of algorithm produced midi files with score based midi files. 
The realization of such a convergence would represent a major step forward in accurately defining 
the parameters of the original performance.