Piano World Home Page Forums Our Most Popular Forums Piano Tuner-Technicians Forum Possible EBVT III implementation on TuneLab

@Kees: Thanks for the TL files. I could try both split-scale and "normal" mode with your numbers next thing.

Bill: My point was just that I'd a like a stretch method that will work for any WT. Personally I use Neidhardt 2, Young 2, Bach-Lehman-reversed and Werckmeister 3 as well as my home grown ones.

Thomas, thanks for the clarification. I think if we get the weigths of various intervals the same my calculation and yours should be exactly identical if I computed with a zero inharmonicity piano. And if IH turns out not to matter in these offsets your calculation is much simpler and more transparant.
If we were to make a small bet I'd put my money on Thomas' method.

Kees

Patrick: The first thing I looked for, and was surprised to see was that no aural tweaking of D5 was needed, it seems. For this particular piano, was that he case?

I tried that today in practice and it sounds very nice indeed.
Theoretically it does not fit within any of the historical WT schemes as all the fifths are of different sizes, but it seems very close to the temperament that is used in the Thomas Church in Leipzig on the Bach organ. That temperament was designed using all historical information to be closest to what Bach probably used. It has only one Pythagorean third on C#, whereas EBWT-18 has 2, on G# and C#.

Very interesting!

Kees

Oddly enough, yes... it is possible that it might be this specific piano, or the scaling. I'd also leave the benefit of a doubt towards my accurateness. If D4 was just a little on the low side, and D5 slightly sharp, it would exlain some of it.I will research it a little, as my own piano is a Yamaha G2. I'll try the numbers on that as soon as I find the time for it!

I tried the EBVTIII 2,1 numbers this evening on a brand new Knabe grand, 6'4". I made a recording, but it didn't turn out very well. To set the ET tuning curve, I set the tuning for 8:4/8:2, and then manually adjusted the mid treble stretch until the twelfths went wide of pure at C7.

Here are my thoughts on it:

These numbers did lead to very nice sounding octaves and fifths, but the fourths were pretty active going up from the temperament. To those who have more experience aurally tuning EBVT III, how much movement can there be in the fourths before it becomes an issue? Some of the fourths were as active as the A4-D5 fourth when correctly placed.

The overall sound seemed to be very nice, and I think I got something that you could call a pipe organ effect in the key of C, but I'm not sure I've ever really heard that, or could pick it out easily anyway.

If anyone else has tried the 2,1 numbers, please let me know how they went.

Regards,

Thomas,

sorry that I didn't write earlier - been busy over here. I had the same feeling about the 2.1, that is, the 4ths seemed a bit busier than I tune aurally. Now, this would mainly be about the initial fourths, when you get up to about A4-D5 I feel it becomes less of an issue.

I have a theory that is just based on empirical research. I've used your original numbers quite a few more times during the week, and what seemed to be random at first boils down to this: I believe that you'd have to be prepared to adjust notes aurally across both breaks - both the tenor and the treble break!

On a Kawai grand, the plain strings above the tenor break were fine, but then I had to adjust their corresponding lower notes forming the fifths (just beneath the break). Then again, on many pianos the last few notes before the treble break seem to need a bit more stretched corresponding higher notes (in the 5ths). That phenomenon viewed from the other side (the last notes before the treble break "taking off" a bit) could possibly be why D5 worked on the Yamaha G2, which has the treble break at D5/D#5. It seems like there are deviations regarding inharmonicity at the treble break, too.

Does this make sense?

Kees,

regarding beats (4ths/5ths/8ths) being less offensive with all strings pulled in, I think it might at least to some extent have to do with the way I tune unisons. When I tune L/R strings to the center strings, I approach them from above and basically never go below the center string.

Alfredo had a great description of this kind of unison tuning, he spoke about "putting smileys on the unison". Isaac wrote about "projecting the tone into the room". It gives an airy feeling to the tone, and seems to disguise even a quite daring stretch of the perfect intervals.

I guess the final limit will always be the inharmonicity model we are using to tune the piano from. I would expect that many pianos have a iH jump or change at the treble break, especially if they don't "dogleg" enough (that may not be the proper word...), in other words, there is a significant difference in speaking lengths on each side of the break. Perhaps the requirements of EBVT III expose errors in the iH model, as the requirements are quite specific.

I wonder how Tunelab reacts when more than six measurements are taken in? If you were to measure iH on each side of each break, would that yield a more accurate measurement, or does the program average out the extra numbers? If Mr. Scott is reading this, perhaps he might chime in. I've noticed that the sample tunings that come with Tune lab can have more than 6 measurements. Or perhaps if Kees knows, he could answer.

I realize that the actual formula used to calculate the iH model is probably very valuable, so I'm not asking for specifics, just a basic idea.

I'll change EBVTIII 2,1, so that it there is an equal compromise between an ET octave and pure fifth. That might calm down some of the fourths above the temperament. Perhaps I will have to add the fourths back into the equation again.

Regards,

This would certainly be most interesting to know

A smooth curve is fitted to the ih data, so taking more measurements does not help. The IH curve is a sum of two exponential functions, as is the tuning curve.

Kees

Pat: I think we're losing focus a bit here. Which offsets work better, mine or Thomas'? Do they have the same problems across the 2 breaks?

Kees

I have full ih data for only two pianos, plotted below with the smooth ih fit. I see a clear possibility to improve across the tenor break with a split scale curve, but nothing that suggests itself at any other breaks.

Even what happens at the tenor break is very messy.

In the past I tried to use the ih for every note in that region to compute just a 6:3 octave tuning and the results were worse than just ignoring the break electronically. Perhaps the ih model per note also breaks down there. Should be easy to figure out. Tunelab's ih model per note is documented in the help menu.


Estonia 190

Heintzmann upright 1901

Kees

Interesting, I didn't realize that the program included that information in the help file.

How much accuracy is needed to properly execute your program Kees? Would a iH model within .1 of the actual iH be sufficient, or higher? Ideally, the end accuracy should be about 1/4 of a cent, the reasonable edge of human hearing on many pianos. Could we possibly set up a more detailed iH model, by measuring four notes within each section? Your program doesn't rely on Tunelab's stock iH model, but could adapt to the higher accuracy needed for a well temperament.

I will try the pianos you privately sent me, and give some feedback on either Monday night or Tuesday evening. If you want, I can sample all the notes, or do the pattern I described above, and we all could compare the results.

you two please carry on... and please throw any aural verification stuff my way.

Like I said earlier, the breaks seem to be throwing us off.

Kees: Haven't had any chance of checking your numbers, but soon i will (tuesday).

Another thing that went through my mind...

Perhaps we should change the partials being used to tune certain areas? Obviously, one would still use the fundamental for the highest notes, but Tune lab defaults to the second partial for the temperament, and fourth partial for the octave below. Perhaps using the third partial all the way up to at least A4. This would set one half of both the fifth (3:2) and fourth (4:3), and would lend itself perhaps to more consistency between the temperament and the area just below.

On the same thought, maybe using a higher partial above A4 as well might help directly above the temperament, as the aural method doesn't really employ the fundamental until F5, the switch to mindless octaves.

Just a thought.

Regards!

My experience is if you use the raw ih data per note, rather than fitting a smooth curve through it the results of a computed tuning based on that are terrible. My guess is that the ih model per note is quite rough (in reality the partials do not form a pattern that can be captured with a 1 parameter model) but if you combine it with smoothing the ih data the errors more or less average out, just as in aural tuning you'd compromise on a weird string.

As far as I know the Verituner is the only device that just measures the partial positions for each note, not trying to fit them into a 1 parameter model) but I don't have one, and its design does not encourage hacking into it if you know what I mean.

Of course if you have all the partials for two notes, trying to tune one to the other, you have to decide what to do with them. Tuning aurally you compromise on what sounds best. Such esthetic value judgments would be very hard to encode. Of course an aural tuner would then also check other intervals.

I'm not sure how far ETD tuning can be pushed beyond where it is now. Interesting topic though!

Kees

Thomas, I'd be eager to try that! Could you make a .tun file where you utilize the 3rd partial a bit more?

The offsets will be identical.

Kees

Yes, as Kees said, my numbers take no account of iH, so the numbers can't change. Simply change the default partial selection of your tuning file. There should be an option to "Edit Partials" or "Edit Table of Partials". Any of the notes can be tuned at different partials, but Tunelab won't automatically select the second partial above C7, though you can manually change to nearly any practical partial.