Disklavier

Good morning, everyone.

In a message dated 12/2/03 11:04:02 AM, re_p_g_c@... writes:

> Why velocities should be not greater than 96 (100)?
> 
This velocity issue comes up every now and then, and I would like to address 
it from another perspective.

Imagine that we are designing a reproducing piano. How would you map the MIDI 
velocity numbers (0-127) to actually hammer velocities?

In the MIDI language, 0=note-off, so you have the velocities of 1-127 with 
which to work.

My initial inclination would be to map 127 to the loudest possible note (i.e. 
fastest possible hammer velocity) and to map 1 to the softest possible note 
(slowest hammer velocity). Then, I would need to make some decisions about 
whether I would map the MIDI values in between 1 and 127 evenly or whether I would 
cluster them in the most used range of hammer velocities (thereby increasing 
the practical resolution of the instrument).

Although the foregoing may sound like a simple concept, further refection 
brings up some interesting issues. What is the fastest possible hammer velocity? 
I can go to the piano and pound away at the loudest volume that I would ever 
employ in a piece by Prokofiev, BUT the fastest velocity that I might generate 
might not be the fastest velocity that someone else might generate.

Given that fact, I should probably alter my theoretical design to create some 
  "headroom" in my scheme. In other words, I should probably map what I think 
is the maximum LIKELY velocity to some number lower that 127 so that I don't 
run out of numbers when I record   someone who actually plays louder them me!

In my own playing on current Disklaviers, I have occasionally generated 
velocities up to about 120, but my loudest playing rarely exceeds 110. Unless the 
piece in question has really thick chords and lots of sustain pedal, velocities 
of 100-105 are usually high enough to achieve fortissimo.

Interestingly, while recording artists for the on-going International 
Piano-e-Competition, I noticed some occasional velocities in the low 120s. When these 
recordings are released, it might be interesting to examine them to see just 
how high the velocities go. I don't think that we should be dissappointed if 
we never find a value of 127. Because of the need for headroom, there is 
probably another pianist, somewhere, who can play even louder than those who 
participated in the current competition.

As for mapping the low end of the velocity scale, we have an even trickier 
issue. Do we want to capture notes with such low velocity that the hammers never 
reach the strings (but whose dampers, of course, do rise from the strings, 
allowing sympathetic vibrations)? I think that we do, in which case we have to 
reserve the lowest velocities for non-speaking notes.

These are tricky issues. I don't have any precise information about how the 
Disklavier engineers came up with their velocity-mapping scheme, but I do know 
that they had to consider these kinds of issues.

Until the Disklavier Pro, Disklaviers have been able to record a wider range 
of dynamics than they have been able to reproduce. With the Pro, recording and 
playback in XP mode are nearly identical. (The Disklavier Pro's XP mode 
actually uses a scale of 0-1023; I believe that it is the same scale of 0-127 but 
with finer resolution between the normal MIDI values. The extra bits of data 
are stored as normally unused MIDI controllers.)

I hope that this has helped the discussion a bit.

Regards,
PianoBench