This whole thread is under 'review' by the PTG because I somehow 'offended' someone, and I did not mean to do so because I kept the name of the app totally confidential. I apologize. I have 'great' relationships with all the underlying people who were quietly behind the scenes collecting the data and even the app rep who appreciated the hard work involved. No offense was taken. The purpose of this thread was to do the hard work to show the data statistically.
I hope the committee does not take it down because many of you have had great responses.
Thank you.
Steve N.
Original Message:
Sent: 02-14-2024 15:32
From: Steven Rosenthal
Subject: Challenging mic vs sensor with independent measurements
Peter, for the most part, my interest in this is not the issues you raise. What piqued my interest was that early on in the discussion of the PianoSens, Steve N. asserted, perhaps in a moment of hyperbole, that in some instances there were differences in frequencies between the string's signals and those emanating from the soundboard that were more than trivial. This could present a problem, perfect unisons of a piano that's not in tune with itself are not of much use. Yesterday Steve N. stated that in fact the frequencies both from the string and the soundboard are indeed the same. Okay, I'll accept that at face value. Problem solved. It was never a challenge of whether the PianoSens "works" or not, I'm sure it does.
For piano tuners, the objective test that we have created a TET are the interval checks. If tuned correctly we will get a smooth progression of beat rates as we move away from the center of the piano. Unlike any readings we get from an ETD regardless of how it gets its input, the interval tests are a direct aural reading of the differences between coincident partials, no algorithms no indirect calculations. It seems to me that that should be the starting point for any evaluation of a piano tuning. Important as unisons are, they are secondary to the tuning of the instrument as a whole. That's why I requested it a month ago, I think it should be checked off the list first, then on to the unisons etc. I understand that the PianoSens doesn't calculated tunings, it's merely an input device, but the difference between reading the string vs. reading the signal from the soundboard I think is enough to verify the tuning first. Had this been done in the beginning I wouldn't have had much to say.
As far as the issue of ascertaining the difference between string and soundboard, which Nathan takes up, it does seem problematic. Perhaps a statistical analysis would give a very educated guess. Still the most direct way would be to evaluate the results of a tuning based upon the string's signal alone using the interval checks, if there is an unevenness, that indicates a disparity if not, then we're golden using that method.
------------------------------
Steven Rosenthal RPT
Honolulu HI
(808) 521-7129
Original Message:
Sent: 02-14-2024 12:46
From: Nathan Monteleone
Subject: Challenging mic vs sensor with independent measurements
Steve R:
Probably the best thing one could do to resolve whether the transduction from string to sound changes any of the partial frequencies, would be to take a _bunch_ of mic readings at different locations and average them (or take the mean, or whatever you decide is more significant -- I haven't thought through it TBH). You could use an offline tool or just take the readings off the app, I kinda doubt it would make a difference when you're averaging a large # of them. You could then compare that average with readings from the string sensor, and see if there's any bias between the two. I certainly don't expect anyone to do that work for us, I'm just offering a potential experiment design.
I agree with your basic question, as I understand it. In my own own words, I would phrase it as "the string sensor has less variance than microphone readings at different locations/times, but is there any bias between the two? And if so, which most closely matches what we actually hear?"
Edit: Apologies if this is already covered in the current data, I am still digesting that.
------------------------------
Nathan Monteleone RPT
Fort Worth TX
(817) 675-9494
nbmont@gmail.com
Original Message:
Sent: 02-13-2024 16:05
From: Steven Rosenthal
Subject: Challenging mic vs sensor with independent measurements
Okay, thank you, the frequencies do not change at all during the transmission through the soundboard. I'll accept your word on that, out of genuine curiosity, how does one verify this?
The reason it is consequential is that the ultimate arbiter is our ears, we don't hear the strings directly we hear what is coming out of the soundboard. I'm sorry that this so irritates you. But thanks for your answer.
------------------------------
Steven Rosenthal RPT
Honolulu HI
(808) 521-7129
Original Message:
Sent: 02-13-2024 15:33
From: Steven Norsworthy
Subject: Challenging mic vs sensor with independent measurements
Frequencies are set into motion as standing waves on the string, as they are the SOURCE. The soundboard and acoustic waves color this with up to 450,000 eigenmodes of spatial color (ref). The frequencies are not changing but the ability to discern and resolve them do change. Spectral analysis at the highest level is needed.
You still miss the most consequential aspect. You have significant variances with the mic. Which mic reading variant do you trust? Why? Is it consequential to YOU? If not, no need to keep debating. Just let it go. Just use your preferred app and your preferred mic and call it a day. I suggest you now let others write their findings, coming up today. Give some room for others, please.
Steve N.
Original Message:
Sent: 2/13/2024 3:28:00 PM
From: Steven Rosenthal
Subject: RE: Challenging mic vs sensor with independent measurements
Steve, variance using the mic aside, was that an answer to my question? This test is using input from different sources, how do you verify that the composition -frequencies- of the two are the same? It seems that this would be essential for cross comparison.
------------------------------
Steven Rosenthal RPT
Honolulu HI
(808) 521-7129
Original Message:
Sent: 02-13-2024 14:53
From: Steven Norsworthy
Subject: Challenging mic vs sensor with independent measurements
Points:
1) Variance using the mic vs invariance using the sensor. It is an already understood phenomenon that moving the mic causes changes in the resolution process of determining an accurate reading of frequencies partials from position to position.
2) The actual reported variance by the app itself vs what the representative says the app reported. The independent study showed large variance in the app itself by moving the mic. The representative cherry picked a few partials per note that lined up with virtually the same readings in the app with the mic vs the sensor to make it look inconsequential. It is already understood that some partials on some notes on some pianos will not vary much in frequency readings with mic positional changes, but some will. Now we have a statistical study that is needed. So an independent did a study and collected the data using the same model piano as the representative used. That study needed to be done in order to more comprehensively show the representative's results were not comprehensive and indeed were very consequential.
3) In other words, if you are selling a $1000 app, it is in your best interest to say that there is no need to buy a $600 sensor in addition because it makes no difference. Well, the independent study shows otherwise. It is up to the reader to determine, from the numbers, whether it is consequential or not. It is not up to an app representative to do that.
4) The representative has dismissed a user for stating how much various apps were seeing significantly reduced jitter, by saying that the user was not experienced with apps and was not using his app. So what then happened? Later today you will see some new reports by this very user who gathered a room full of expert app users and evaluated this on a grand piano. Stay tuned (pun intended).
Steve N.
------------------------------
Steven Norsworthy
Cardiff By The Sea CA
(619) 964-0101
Original Message:
Sent: 02-13-2024 14:33
From: Steven Rosenthal
Subject: Challenging mic vs sensor with independent measurements
Steve, the PianoSens is measuring frequencies emanating from the string while the mic is measuring frequencies emanating from the soundboard, how do you verify that the two are identical?
------------------------------
Steven Rosenthal RPT
Honolulu HI
(808) 521-7129
Original Message:
Sent: 02-13-2024 09:09
From: Steven Norsworthy
Subject: Challenging mic vs sensor with independent measurements
MATLAB is the international standard, de facto
Original Message:
Sent: 2/13/2024 8:58:00 AM
From: Parker Leigh
Subject: RE: Challenging mic vs sensor with independent measurements
Have you ever considered Comsol Multiphysics as a platform for measurement?
------------------------------
Parker Leigh RPT
Winchester VA
(540) 722-3865
Original Message:
Sent: 02-13-2024 04:44
From: Steven Norsworthy
Subject: Challenging mic vs sensor with independent measurements
No app can self evaluate. The rep for the app cannot use the app to self evaluate. The data collected by the app must be subject to a higher authority measurement like Matlab. It is the only way to get an independent evaluation at a much higher level of accuracy. The app cannot do that.
An electronic string sensor makes most of the apps tested so far reduce this variance as the app reports, but then the app itself has its own variance due to limited FFT length and window size and limited inter-bin measurement.
The particular app is not named because most of the tuning apps have a similar mic placement variance.
An app representative told me he saw no significant difference (?) in measuring his Yamaha C7 with the mic and then the sensor. That was simply uncorrelated with vast data I have already collected but not using his app. It also says, effectively, that his app is completely insensitive to mic placement variation, a statement that does not hold up, as we shall see.
I don't want to pit app against app. But I had to clear the air with this issue. It left me with no choice, but to do independently have someone conduct an experiment on a Yamaha C7 with his app.
I had another well respected professional use this same app and go to a Yamaha C7 and take comprehensive measurements. He used 9 different mic positions per note and measured C1, C2, C3, C4, C5, C6, C7 and did a screen video recording of what App X was reporting and captured the .wav file audio so that I could then take the audio into Matlab to see what it actually is doing, and correlate it to that app.
The app in this case could not resolve the variations in the spectra. The mic variance of the 4th harmonic, for example, of C3 was off by almost 1 cent. I typically looked at the first 5 harmonics in Matlab but have all the higher ones of course available. The C6 2nd harmonic was 9 cents off, yes that much. The C7 2nd harmonic was off by 19 (nineteen) cents. The upper register of the piano is literally filled with all kinds of false beats that show up in the spectra in matlab with great precision. We hear this junk with our ears also. The sensor is able to pick out the dominant one as previously shown in my video presentations.
The app reported the following mic spreads. The actual was far worse with the same data evaluated in Matlab. In Matlab, the variances and ranges were significantly worse and wider than what the app reported, but the app had ranges as bad as 2 cents.
The ranges as reported in the app were (C1 through C7);
1.1, 0.66, 2.34, 0.46, 0.21, 033, 0.61, in cents!
In Matlab the ranges were wider, depending on the length of FFT in the measurement. Since the FFT length and window functions and inter-bin calculations of App X were not known, there was no way of correlating in a general sense. I typically used 500 msec or 1 sec for the FFT length in Matlab. Most apps cannot handle this long so the variances are greater since the timeframes of the windows are less.
Remember the sensor does not vary and this 'invariance' has previously been presented with extensive tests.
Here are the details from strike to strike as reported by his app. Again, these are not my readings, but the readings of the app in question, and these were not done by me, but by an independent party.
These are the mic measurements in cents offset from zero reported by the app:
C1 = 1.1500 0.0500 0.1100 0.7000 0.5400 0.7000 0.6400 0.1900 0.2800
Range = 1.1 cents
C2 = -0.8200 -0.7300 -0.8500 -0.4300 -0.6200 -0.1900 -0.7100 -0.6600 -0.5700
Range = 0.66 cents
C3 = 2.2500 0.3500 0.7300 0.4000 0.2400 0.3300 0.4200 -0.0900 0.0800
Range = 2.34 cents
C4 = -0.6100 -0.6800 -0.5500 -0.8400 -0.6600 -0.7400 -0.8900 -1.0100 -0.9700
Range = 0.46 cents
C5 = 0.1400 0.2300 0.2200 0.2700 0.0600 0.1000 0.1700 0.1900 0.2600
Range = 0.21 cents
C6 = -0.2900 -0.2600 -0.1700 -0.2400 -0.3000 -0.5000 -0.2600 -0.3700 -0.2200
Range = 0.33 cents
C7 = 0.9800 0.9100 0.5000 0.7600 0.8500 0.7600 0.4700 0.6400 1.0800
Range = 0.61 cents
More later, with spectral pictures to come.
Best to all,
Steve
------------------------------
Steven Norsworthy
Cardiff By The Sea CA
(619) 964-0101
------------------------------