Piano World Home Page Forums Our Most Popular Forums Piano Tuner-Technicians Forum Using Easy Piano Tuner

Following up on the overpull percentage conversation, I spent some time over the past couple of weeks collecting and analyzing a bunch of overpull data on several different pianos in an effort to develop a slightly more sophisticated model for overpull. The method currently in the app basically uses a linearly increasing overpull where the percent overpull is adjusted slightly based on the scale of the piano (estimated from inharmonicity) and attenuated at the ends of the piano. What I've found recently is that the overpull profile seems to be quite different for grands vs uprights.

Here, for example, is the profile for a grand piano:

It fits the old linear model fairly well.

Here now is some data I took on a studio upright piano:

There is a dip in the overpull through the midsection and then a large jump as you cross the strut into the treble. The overpull percentage still varies with the scale of the piano. Note that the actual overpull percentages are different than you're probably used to seeing (eg. 50% in the treble). That is because we are pre-measuring all the sample notes before beginning the pitch raise, instead of measuring them as you go.

Anyway I wanted to ask if this difference between grands and uprights is consistent with others' observations. I'm hoping to get something in place in the next several weeks. I still need to tweak the model, write the code, get it into the app, and test it.

Yes, that is similar to what I see when approaching uprights above the strut - different models seem to be a bit random approaching the strut, some need a bit more, some need a slight taper. But excepting Steinway uprights, almost everything I've tuned needs a bigger boost in percentage overpull going above the strut.

For my lever technique, most pianos get a 17% in the middle(some 23%). I don't often see the sag that you noticed, but I'm calculating percentages as I tune, so results will differ...

Above the strut jumps to 28% for most, though some stay at 23% for a few notes - jumping up to 33% and some even 38% in that crazy section, then tapering back down.

Ron Koval

I have been 'mirroring' all of my tunings with EPT (just let it run as I tune) - it has been really interesting to watch the inharmonicity graph. Not sure how to post pictures, I'm sure other techs would like to the those graphs for different common models of pianos! It sure gives a nice visualization of why some models are a little more difficult to tune than others...

Ron Koval

Re. the sag in the midsection: I think it could be modeled just as easily by a straight level line through the midsection with a jump at the treble break, but with a few notes around the tenor break that need a little more overpull. In many of my samples it seems to be the notes closest to the tenor break that need the extra boost in overpull. I suspect this has something to do with the lower tension, foreshortened strings at the bottom of the tenor (the ones that also move most with humidity) and possibly edge effects of being on the end of the bridge. Another possibility is that in addition to precalculating the overpull, I am tuning in a different order than I think other piano tuning apps recommend. I start at the bottom of the tenor and tune to the top, and then tune the bass section last from the break downward. (I remember at least one app saying to start at note #1 and tune upward, but that has never sat quite right with me, especially if you're measuring inharmonicity as you go.)

On the inharmonicity graph, I do the same thing, and it's made me think about pianos differently, especially the spinets with unwound bichords. Actually, when collecting recordings for these pitch raise calculations, I forgot on some of the pianos to make a note of where the lowest unwound string was. I was able to reconstruct that by quickly running the notes around the break through EPT and looking for the jump in inharmonicity.

I'm not sure of the best way to do images here, but I upload mine to https://i.imgur.com and then paste the "direct link" here between "img" tags like this:

Could the developer or anyone familiar with EPT elaborate on the two other screens ? How to analyze the two other screens ?

Thank you.

FY

PS. I read the very short explanation at the support site, but it doesn't say how to analyse the information.

I really like the information found on the inharmonicity screen. It gives you a look of each string's partial ladder boiled down to an inharmonicity constant. The graph shows an idealized curve and the measurements are plotted over the top. It lets you see the stringing breaks, but more importantly to high-level ETD users, It gives some control over the quality of the data used for the calculation. If you happen to see a string, or many strings "out of place", you can go back and keep sampling that note until you get consistent results.

I would think to achieve a higher-level tuning with this software, it would be best to review, remeasure and then lock the tuning calculation once everything has been measured.

Ron Koval

I have never used the overpull function on TuneLab or Simple Piano Tuner. Since each of the duple and triple choir strings have different non speaking lengths, how can a single overpull percentage calculation for a note be accurate?

Ian

Is there a way to view the numeric data...inharmonicity constant, tuning curve offsets, partial used?

Yes, kind of. You can extract data for the measured inharmonicity, and tuning curve offsets, but it's not terribly straightforward, as that's not really the point of the app. I've written down some instructions here: http://my.ptg.org/communities/commu...r#bmf3db84ea-ee53-417a-8855-846b8e6f0d91 (gasp what a URL)

If the link doesn't take you directly to the post scroll down to post #10.

Oops! Sorry I meant to write Easy Piano Tuner.

Ian

It's not. The overpull calculation is a big estimation intended to get you "close enough" for a quick first-pass tuning. That said, effects due to the difference in non-speaking lengths are going to be small compared to the other factors that make overpull necessary. (This isn't based on any actual data, just my own observation.)

A fun exercise might be to try to predict whether you'd need more or less overpull for a string with a longer non-speaking length compared to its (shorter) neighbors. I would predict less.

The main purpose of all 3 screens is to be a sanity check for tuners.

The tuning curve screen shows you what the calculated stretched tuning looks like compared to (flat) equal temperament, and gives you a rough picture of how your tuning lines up to the targets. It's not accurate or fast enough to actually use as a tuning aid, it just gives you a quick birds-eye view.

The inharmonicity screen gives you a view of the measured data that is used to calculate the tuning. It basically makes the program less of a black box. Ron's advice for using that is excellent. Ignoring it completely won't give you a bad tuning, but checking it to fix outliers and missing data points can give you a slightly better tuning.

The frequency spectrum screen isn't helpful for tuning at all but it can give you some insight into other things. I use it occasionally to confirm things I notice first with my ears, like an annoyingly strong nth harmonic. I've also found it's useful for explaining the harmonic series to clients.

Anthony,
During my normal (not overpull) tuning using Easy Piano Tuner, I am very aware that in using my carbon fibre lever I notice tension differences between each of the three choir strings. Also the range of NSL tension that remains stable is different for each choir string. The shortest NSL has the least range and requires the most careful setting. This is why I suspected that a One-For-All over-pull function would not be so accurate. Perhaps I misunderstood the tuners here who seemed to use the over-pull function for a single pass tuning.
Ian

We may be using different terminology. All 3 strings of a unison should have exactly the same "tension" when they are in tune, regardless of the non-speaking length, assuming the speaking length is the same across the unison (it should be in a good piano). That said, there would definitely be irregularities in friction and tuning pin tightness that would cause the force you feel in the lever to be different for strings in the unison. I suspect that's not what you're talking about though. What you are describing I think is the difference in "sensitivity" between the different strings of the unison: if you turn two pins exactly the same amount you will get a larger pitch change from the string with the shorter non-speaking length. Or put another way, you have to turn the pin more for longer strings than you do for shorter strings, even though the speaking length is the same. This is because raising the pitch stretches the steel itself and in longer strings there's more steel to stretch, so you have to tune the pin further to get an equivalent change in tension to you would you'd get in a short string.

I don't doubt that some tuners in some situations use just one pass and fine tune to the pitch raise targets. I've also heard that cybertuner has a fine tuning mode that adds just a fraction of the normal pitch raise overpull, so I imagine some tuners use that for one-pass tunings. I personally prefer doing a two-pass tuning. Even if the piano is close enough to skip the pitch raise altogether I still start with a quick (<5min) pass to touch up any outliers.

I don't know of any tuning program that compensates for differences in non-speaking length.

I continue my journey with Easy Tuner. It's comfortable! This is high quality! But I sometimes find tools that seem strange. And these strangenesses are visible in the behavior of the phase display. The bands of overtones are moving in different directions:



About this speech came a little higher in the forum. But I still do not understand this behavior. Also, in the range where such strings are present, there are inaccuracies in tuning. If you move parallel intervals in this range - the intervals change regularly and sequentially, but the harmonic chords in some keys may sound false.

 It turns out that when measuring the non-harmonicity, the note is recorded by some idealized values ​​of the overtone frequency. At the same time, there is an obvious slight discrepancy with the actual frequency of some overtones. When calculating the setting and during the configuration, is this idealized value used? Then it leads to inaccuracies, mistakes and oddities. I was impressed by the Entropy Tuner that that tuner measured the actual frequencies. And it was the actual frequencies used for the calculations.
 This reminds me of a situation where you have some imaginary space, for example 5 meters wide, which you need to cover with boards 1 meter wide. The boards come, but the width of some is 0.95 meters, 0.93 meters, 1.02 meters, etc. for example. (hehe in Russia is so often)). It is clear that nothing worthwhile will come of this - there will be gaps and discrepancies. The rhythm will be disturbed. But in this case, we should start from these unevennesses, and not from idealized meanings. - An example came to my mind))
 I think that Easy Tuner should be aware of these irregularities. Should keep the corrections for uneven overtones and take these values ​​when calculating the beats in the intervals. (and at the same time to stop the divergence of the lines in different directions in the measuring mode).
 Although, on the other hand, we can detect "strange" strings and thus have the opportunity to find alternative solutions for specific strings.
 Also there is a suggestion to make a demonstration of the overtone volume in the phase display using the thickness of the lines (or brightness) (changing the width of the lines depending on the height is good!)
And another question - how does the tuner react to the presence of longitudinal modes? Some strings can have quite bright and fake frequencies of internal beating strings. Do the longitudinal modes lead to an inaccurate calculation of the nonharmonicity of the string?

EPT looks like another interesting program.

How many notes does EPT measure, to generate a tuning curve? Is it like the Cs and Fs in six octaves, like in Tunelab? Or is it every single note, like
in Dirk's or Entropy? Can the number of measured notes be modified by the user?

Is there a split-scale mode in EPT, for spinets, like Tunelab has?

Sorry if these questions were answered earlier.....

Does anyone sometimes see the needle flicker uncontrollably and widely when it is listening but there is no note being struck?

I cannot figure out why it does this on my Android LG G4 phone. No other apps are running.

Ian

I doubt if any of the longitudinal wave energy in most piano strings is going to show up in the display. Maybe in the first 10-15 notes you may "find" it somewhere up in the range of the 20th partial.

Your phone (likely) uses an automatic gain control (AGC) on the microphone preamp. Just about every modern audio recording device has this dubious ‘feature’. It adjusts the amplifier’s gain higher for weak signals and lower on strong signals to prevent distortion. The rate at which the gain changes is usually very slow (more than two seconds). In your case, if no note is being tuned, the AGC goes to maximum sensitivity, so it is picking up any noise in the room as well as audio frequency EMR, possibly not detectable by you, but is detected by the mic. This is normal.

Alex,
Thank you for the perceptive feedback.

You are correct about the phase display bands moving in different directions. It usually happens in the tenor section near the "break" and it is caused by a small discrepancy between the inharmonicity EPT is using for the note and the actual inharmonicity of the string. EPT, as you know, calculates inharmonicity for each individual string, and also calculates a best fit curve to model the inharmonicity of the entire piano which can be used to estimate the inharmonicity of strings that haven't been measured yet. The inharmonicity values are continually refined as you take longer measurements, and the changes to the inharmonicity values become progressively smaller the longer you measure. However EPT still doesn't directly use these raw inharmonicity values to calculate the tuning. It uses a weighted average of those and the calculated best-fit curve. I realize that probably sounds like a bad idea, to use averaged values instead of actual values, but it actually helps to prevent disruption to the tuning curve from bad inharmonicity measurements. Also, the actual inharmonicity values are given more "weight" in the average with the best-fit curve the longer they have been measured. So for a note that you have measured for only 1 second, the inharmonicity value used will be mostly determined by the best-fit curve; but for a note that has been measured 20 seconds, the inharmonicity used will be much closer to the real inharmonicity. One thing I could do to give you more control over this would be to add an option to adjust the "weight" given to the average best-fit tuning curve. That way you could choose to have it use the actual inharmonicity values exclusively after they have been measured.

The other issue with the strobes going in different directions is that the actual spacing between frequencies is uneven and doesn't perfectly fit the mathematical inharmonicity model. Early in the development I tried adding a correction factor for this, but it ended up creating more problems than it solved, letting noise corrupt the data and making messy and jagged tuning curves that sounded bad in some areas.

As things currently stand the best thing to do when you have strobe rings moving in different directions is to tune so that the middle strobes are stationary and the outermost and innermost rings move in opposite directions.

Your suggestion to change the brightness of the strobe rings based on the loudness of the harmonic is something that I definitely plan on doing in the future. (I don't have an ETA on it unfortunately, but it's on my list of things to do.)

Re: longitudinal modes, I have logic that should mostly eliminate interference from longitudinal modes. If we detect two frequencies sufficiently close together (like a real harmonic and a longitudinal mode) we throw both of them away and don't use either of them for the inharmonicity calculation.

As many as you want. At least 3.


I will usually begin my tuning by taking some quick samples like in TuneLab. I usually use a few octaves of A's and D#'s, though C's and F's work as well. Some users will "lock" the inharmonicity (using the "ear" button on the right) after taking these measurements and tune the piano, making it function very similarly to TuneLab. Others (including me) leave the inharmonicity measurements unlocked for the entire tuning, meaning that the program will constantly be measuring inharmonicity and slightly updating the tuning curve as you go. If you are tuning the piano in two passes (pitch raise, tuning) this isn't a problem as any later changes to the tuning curve will be very small. I basically tune all my pianos in two passes: even if a pitch raise isn't necessary I still run through all the notes to touch up any outliers before doing the actual tuning.


Hmm. I'm not sure what that is. I don't think so.