Piano World Home Page Forums Our Most Popular Forums Piano Tuner-Technicians Forum Tunelab EBVT plug-in

Hi all,

I implemented an EBVT plugin for tunelab. You create a tunelab .tun file with the inharmonicity measurements of, say, A1A2A3A4A5A6, it reads that and computes the offsets for EBVT by essentially going through Bill's extensive recipe including the details of the octaves. It does this by using a standard tunelab inharmonicity curve fitted to the measurements. When the calculation is done the results are saved in a tunelab .tun file which has all the custom offsets in it for that particular piano. You open it and there you go.

An example of the resulting tuning curve for my upright is below.

I'd be happy to share the program, but I doubt many tuners will have MATLAB in which it's written.

If you want to try it out you can email me (kvandoel@shaw.ca ) your tunelab file with the inharmonicity measurements of the piano and I can generate the custom .tun file with the offsets for you. I would be very interested to see how it measures up to aural tuning. My aural skills are not yet refined up to the level where I can assess this myself.



Kees

I don't have MATLAB, but can you explain grosso modo what you do to tune stretched octaves of different sizes and mindless octaves?

I suppose you left untouched the offsets for the temperament octave, wich is A3-A4 in TuneLab, not F3-F4 as in aural EBVT III tuning.

Also how do you (if you do) tune A3-A4 as a 6:3 octave and F3-F4 as a 4:2 octave?

Rafael, I just compute the frequencies and per Bill's instructions, and that gives offsets from ET and I dump that in a tunelab file. After that tunelab doesn't do anything except read the custom offsets and assist me in tuning that.

Are you asking how I computed the frequencies?

Kees

No.

The computed offsets of notes are not enough to tune EBVT III accordingly to Bill's aural tuning instructions.

In fact the EBVT III figures are already available within the temperaments provided by Tunelab. You only have to choose this temperament.

I thought you were talking about a plug in to modify the way Tunelab works with these figures.

Let's take an example:

In the second step of Bill's sequence you must tune A3 as a 6:3 type octave from A4.

In order to achieve that you must tweak the tuning of A3 provided by Tunelab wich gives an offset of 0.0 cents for A3. If you let it as it is it won't give you a 6:3 octave but a comprommise between 6:3 and 4:2.

To get an accurate 6:3 type octave what we do usually is a direct interval tuning, setting Tunelab to read the 3rd partial of A4 and then tuning A3 to that setting, so the 6th partial of A3 will be exactly the same that the 3rd partial of A4.

In this example things are easy, but when you have to tune F3 to beat at 6 bps with A3 things are not so easy.

They are not easy neither when having to tune D4 to have equal beating between G3-D4 fifth and A3-D4 fourth.

And things get less obvious when you have to tune an octave which must satisfy a good sounding fith and a good soundig fourth. Taking into account that some fifths/fourths are pure and other are tempered each one by differents amounts of tempering. What gives you octaves of different size for each note in the temperament span.

When you give Tunelab the offset of each note in the temperament setting, Tunelab calculates a tuning curve that evens things to a smooth progression of beats in the choosen intervals, so subtilities as equal beating intervals, pure fifths/fourths and other minuties are exchanged in Tunelab by reasonable close computed approximations.

The same is true for the stretching of the tuning. Bill directions individually set octave stretching of each octave, which will be totally diferent for one octave to the one a semitone higher.

Tunelab calculates a tuning curve with a smooth variation of the stretch applied all along the scale, so the stretch applied to each octave is not individualized but is instead uniformized.

That makes the computed tuning to sound substantially different from the aural tuning. Even if the offsets of each note and the overall stretching are apparently the same.

In other words: Tunelab gives a smooth curbe, while aural tuning gives individually apparently random spots.

And to this day, there is no tuning software able to tune EBVT III exactly as it is tuned by ear.

Maybe Bill can give a clearer explanation of this.

Rafael:

You completely misunderstand what I have done. Have you not looked at my tuning curve? I compute the offset for every pitch (88 of them). That's all. Tunelab doesn't do anything, it has a tuning curve that is zero everywhere and it adds my custom offsets to every note. It reads that from a custom .tun file I create in MATLAB.

Maybe I'll post an example of some of the calculations involved in the "Advanced tuning math" thread later.

Kees

I was writing at the same time you were. Please read my post above.

I have MATLAB, and I've emailed you a .tun file. Can you send me the program?

Ok I got the picture.

In order to translate Bill's instructions into offsets given to Tunelab you have to measure each one of the partials involved in the intervals used by those instructions.

Do you measure all of them? And even if so how do you tell Tunelab wich partial to use in each case?

If not, then how do you deal with iH?

Rafael, can I suggest you read my orginal post, as the answer to your question about iH is there.
In the tunelab file format, after the note is a number which is the partial to use. This has nothing to do with the tuning curve really it's just the partial used when actually tuning.

Kees

Sorry I don't know the format used by Tunelab files.

Also I guess I must see the calculations you do to get the offset of each note.

But there is one thing that bothers me: Do you read directly with tunelab all the partials needed for each note?

Let me explain.

1. Tune A4.1 (first partial of A4) to 440 hz.

2. Measure A4.3 (the actual frequence of 3rd partial of A4) and tune A3.6 to that frequence.

3. Measure A3.4 subtract 6 hz and tune F3.5 to the result.

4. Measure F3.3 and tune C4.2 to that value.

5. Measure F3.4 and tune F4.2 to that value.

6. Measure C4.5 add 6 hz and tune E4.4 to the result.

...etc


So for this 6 steps you need the offsets of:

A4.3, A3.4, F3.3, F3.4, C4.5


As you can see for F3 you need to know the offsets of two partials the 3rd and the 4th.

The 3rd partial of F3 is used to tune C4.
And the 4th partial of F3 is used to tune F4.

But Tunelab has only one offset for each note. So you can't store both offsets for F3, how do you handle this?


Do you individually measure each needed partial with Tunelab and then store it in MATLAB to make calculations?

Rafael:
If you know the inharmonicity constant of a note you can compute all its partials.
Kees

Yes, but by doing so you let the software take over the "aural" tuning.



Yes, you compute the 88 offsets but you let tunelab figure out the partials needed to make the computation by taking its iH constants, which are in turn estimated by the program based on sample measured notes (say A1, A2, A3, A4, A5, A6) and which are in turn used in an empirical formula adopted only by ROBERT SCOTT. Other software will use different formulas.

So even if the curve is "zero everywhere" and you put all the offsets, the substantial iH data is estimated not measured.

I am afraid than the tuning you can get from this approach will differ a lot from an aural EBVT III tuning.

If you were to measure each real needed partial and use this data in Matlab or Excell or whatever calculation software you prefer, then you'll have a more accurate tuning which will be really close to an aural tuning.

You have to measure approximately 3 partials of 76 notes (you don't need the iH information for the top octave).

In fact when you measure a note with tunelab the program shows the offsets of several (8?) partials of this note at once. But I don't know a way of storing these data for later use, only the iH constant computed for this note is saved by the program.

It will be a only once job to take iH data for each different piano. Maybe in some special cases it will be worth the work involved.

Assume GRANDPIANOMAN spends a whole sunday to take the data of his M&H BB. Then he doesn´t need Bill Bremmer to come in to tune his piano, isn´t it nice?

Of course if you are to tune a different piano each time then it won't be worth the effort to take all this information.

Excuse me if I seem to be knocker, but I've worked a lot thinking about how to do exactly what you said. And when I saw your post I was wondering if you have finally found a shortcut to what I have found my self.




I have another question:

Once you have the 88 offsets computed by Matlab how do you enter them into a tunelab tuning file? Do you do it by hand, one note at a time? Or do you convert your Matlab file to tunelab tuning file format?

Rafael:

You are right, the tuning will be only as good as the iH model. I suggest read the "advanced tuning math" thread where this was all discussed in detail.

However it will be as close to a "perfect" aural EBVT as using tunelab for ET will be to a "perfect" aural ET. How that will work out in practice I don't know, but at least I got the pipe organ effect.

A few aural EBVT tuners have emailed me their tunelab files, I am curious to see how good the results are in practice. I hope they get back to me. Of course all you have to do is tune EBVT aurally and then check with what tunelab dictates.

Anyways, feel free to email me your tunelab file and you can try it out yourself. And of course I don't enter anything by hand! I just run the program and it spits out the needed file.

Kees

I've made some tests, but the data I've got from tunelab and verituner are not consistent.

If you measure a given note five times, you get five different readings, which are not at all close to each other. It will give you different values depending on how hard you play each key and even if you play with a consistent blow you can get random values. Sometimes you get even negative offsets, which means the partial's frequence is lower than its theoretical value. I don't know if it is wrong, but at least it is surprising, and makes me feel uncomfortable with the reliability of this measured iH. It gives you different data from one string to another of the same unison. For some strings there are holes in the partial offsets, that is you don't get the offset of some given partials no matter how many times and no matter how hard or soft you play the key. So there will be unavailable partials for some calculations, so you can not tune some notes! Unless you make a multipartial computation in your tuning as Veritune claims to do.

Believe me it isn´t as easy as it seems!

I really appreciate your offer, I will send you my file for a studio Petrof P117K1 and I will analyse the tunelab tuning from an aural point of view and I will make my comments. (During next week)

I think it will be by these kind of cooperative work that piano tuning is evolving each day, maybe a little step but in the forward direction! Imagine what piano tuning will be in say ten or 15 years from now!

Thank you very much.

Rafael:

I don't know what you are doing exactly but I have never gotten wildly different readings for the iH of a note in tunelab. They are always close.

When you measure iH in tunelab it measures up to 8 partials and fits that data to a modified Young model. Within that model all partials are then available. If has nothing to do with partial offsets and partial selection during tuning.

To measure all partials on the fly without fitting them to a model is undoubtedly better. Maybe if I get a Verituner for Christmas I can write a plug-in for that device.

Kees

I have a Verituner and I don't know how to measure and capture all the partials of a note.

In measuring mode you have to select which partial you want to measure, then you measure it, one by one for each note. And as Tunelab it stores only one partial for each note.

I don't know why. Both programs have all the data available when "listening" to a note, but they only store one partial per note.

Maybe Dave Carpenter and Robert Scott are watching us, and will do something about it!

Maybe my piano is inconsistent and not Verituner nor Tunelab.

I wish I could measure a steinway D!

BTW, what piano have you measured?

I'll give you some readings I use to get.

Interesting...given the inharmonicity constants

from my .tun file, the program spits out a custom offset for each partial, using the same partials as the source file (or default; haven't looked at the program yet).