Crazy Entertainment Idea...

[Doc]

I think software would be the only practical option. A microphone and notch filters asigned to logic gates through software.

[Mr.Coffee]

I think software would be the only practical option. A microphone and notch filters asigned to logic gates through software.

notch filters? are you talking about isolating notes? because that wouldn't give you individual notes, unless you had something like melodyne running real-time. You could run a simple FFT to replicate velocity sensitivity though.

[Doc]

I used the term notch filter a little bit loosely there (as Sheriff John Bunnell would say; "like some kind of renegade forum user with trigger happy keyboard fingers, and a disregard forum protocol, typing his way recklessly to date with oblivion" ).

I was of course talking about software performing digital sample recognition thus acting in the way of an analogue notch filter.

[ImageLight]

There are MIDI piano controllers available that would have velocity and after-touch built in. You'd just have to specify where and what those controllers do in software.

[Mr.Coffee]

The issue isn't finding a controller with veolcity capabilities, the issue is adapting something to generate a veolicty sensitive signal. Since this is usually a part of the key mechanism for each key, a generalized solution is in order.

though you're absolutely right, many midi keyboards are available. the Yamaha midi-player piano comes to mind, maybe rigging something like that in reverse?

[Pianoman]

I understand what you guys are saying, but unfortunately there is no keyboard that can replicate the feel and sound of a real piano. These traits are necessary when you are playing extremely difficult technical music. Is there a way to mic the piano and convert the sound waves to an optical signal?

[heroic]

You can use an FFT to extract individual notes from an audio recording of a close-miked piano. You even get velocity sense from analyzing the phase space of the signal. It's been done (notably, by people extracting scores from old recordings). All the information is there; if you can *hear it*, it *has* to be there, after all.

Remember that the human ear works in the frequency-intensity domain, and the nervous system works in the time domain. Everything you hear is effectively FFTed by the hair cells in your cochlea. Frequencies stimulate one particular hair cell (or a number of them), and the amount of energy in that one frequency determines how frequently the nerve it's attached to fires.

I would start by sampling at a high frequency, FFTing, and plotting it as a graph of frequency-intensity, to see if you can get sharp enough peaks (try a Hanning window to localize them). You should be good with say n=256.

The higher the sample frequency, the shorter the period of time represented by each FFT block will be; 256 samples at 48 kHz is 1/187.5th of a second. You only need to generate one frame of laser output every 50th of a second or so, so this means you can do velocity analysis simply by computing the rate of change of each frequency bin. No need to use wavelet or even phase analysis.

A reasonably fast computer will do it in realtime; then, turning it into a stream of samples to feed to something like an EasyLase or suchlike is relatively simple. I would suggest that you can do this in one of the many audio programming environments that are available now.

[Pianoman]

That makes sense kind of, but I'm having trouble understanding. What exactly does FFT stand for? how would one go about sampling FFT and plotting it as a graph of frequency/ intensity? if the peaks on the graph are not sharp. what does this mean, a less intense sound? I will definitely need some help with this, as it is beyond my realm of understanging or knowledge. Once I "see" it like an example I'll be much more likely to understand, I'm a visual learner.

[heroic]

FFT is a Fast Fourier Transform, and is a standard mathematical technique for taking a signal that's in the time-intensity domain and turning it into the frequency-intensity domain.

Here is a nice page that explains it better than I. http://en.wikipedia.org/wiki/Fast_Fourier_transform
And here's a program with a free trial that you could try playing with: http://www.sigview.com/
Fortunately the FFT algorithm is published widely and there are many FFT implementations that coders can help you out with.
I have a couple myself I'd be happy to post.

Basically, FFT is what spectrum analyzers use. Except in this case you are using a spectrum analyzer with hundreds or thousands of bands!

For your project to work, you need to be able to tell one note on the piano apart from all the others. Since when you hit one key, mostly one wire is hit by a hammer, but all the others vibrate a little in sympathy, the signal is not clear (as it would be, say, if you were playing an Ondes Martenot or something like that) and it may be a little tricky to get the exact 1:1 relationship you are looking for.

I actually solved this problem many years ago when I was designing a device to assist bicyclists in tuning the tension of the spokes on their wheels; it plucked each spoke in turn and used a microphone and FFT to determine the tension of each spoke. What you are trying to do, rather, is to hit a random wire and determine from the sound which wire you hit.

It shouldn't be *too* hard, but to get enough margin you may need to sample at 96 kHz (not hard with modern equipment) and accept any

[buffo]

Well, so long as we're dreaming about this project and money is clearly no object , why not purchase a real piano that has velocity sensitive midi output built in?

No, I'm not talking about an advanced keyboard, I'm talking about a real piano with real action that ALSO has built-in electronics to pick up each and every key that is played and convert it to velocity-sensitive midi data.

I looked at one of these pianos about 15 years ago when I was seriously considering getting back into piano. And it turns out that adding the midi-out option to a standard piano isn't all that expensive. (On the Yamaha I was looking at, it added around $1500 to the price.) Granted, if you want something like a baby grand, you're going to pay more for the midi option, but then again, you'll get a bunch more features with it too.

Adam