PianoDisc 228 CFX Attack Time Delay
For Peabody Computer Music's PianoDisc Piano

Craig Stuart Sapp <craigsapp@jhu.edu>
20 November 2002

The following picture is a three-dimensional plot of the timing delay for every possible MIDI note and attack velocity. The left-right axis is the MIDI attack velocity, going from 1 (soft) on the left to 127 (loud) on the right. The MIDI notes are arranged on the near-far axis with the lowest notes being in the lower left side of the picture and the highest notes being in the upper left side of the picture. The timing delays are displayed on the vertical axis. The delay is the time it takes between sending out a MIDI message from a computer and the time that the same note returns back to the computer from the PianoDisc piano.

Note that louder pitches take less time to return than softer pitches. This is because of the physical time it takes the piano levers to be moved. Softer sounds require the levers be moved slowly, and loud sounds require the levers to be moved quickly.

At loud attacks, the timing delay is around 50 milliseconds for all pitches, although 45 milliseconds is about the fastest possible round trip time for very high notes. The cause of the 50 millisecond delay is that the software for playing the piano buffers the MIDI data to correct for simultaneity problems between loud and soft notes received at the same time.

This picture was created by averaging 200,000 random pitch/velocity specified notes over the course of about 5 hours. Each pitch/velocity point was played about 10 times. These values were averaged together to remove any timing glitches in the measuring process.

Here is the same data as given in the 3D plot above. The pitches are along the y-axis, and the attack velocities are along the x-axis. Note the loss of feedback at low velocities. This picture describes the quiest notes which can reliably be played on the piano for each pitch. Note that the mid range is particularly stiff, and it requires a higher attack velocity to play any notes in the midrange. This has been adjusted by Michael by putting teflon powder on the leather hammer knuckles. I will be doing another velocity/pitch timing analysis and see how the pictures has changed.

Notice the sharp change is sensitivity between MIDI notes 47 and 48. This is a breaking point between registers on the piano, but was particularly large on this piano. Note 48 was balanced at 74 grams, while note 47 balanced at 63 grams. Ideally regulated grand pianos should be 54 grams at the low end and smoothly transition to about 47 grams in the high registers.

It is interesting to note the probably change between white and black notes in the picture. There are jagged delay responses which alternate high/low for every other key, in general. Here is the raw timing data used to generate the picture below: timearray.h.

The following plot is only a style variation of the previous picture:

The next sequence of plots displays the amount of time it takes for a round trip to the computer for different octaves of the note "C". If the timing value is zero in the plot, this means that no notes were ever returned to the computer for that attack velocity. This implies that notes played with attack velocities around those values will be unreliably played. You should avoid those regions if you want pitches to always sound when playing notes from the computer. This data will change since the piano has been regulated in the midrange.

Here is a view of the time dela versus attack velocity.  This pictures shows the worst possible delay for all notes. The x-axis is the attack velocity from 0 to 127, and the y-axis is the amount of delay in milliseconds