|
Welcome to the Piano World Piano Forums Over 3 million posts about pianos, digital pianos, and all types of keyboard instruments. Over 100,000 members from around the world.
Join the World's Largest Community of Piano Lovers
(it's free)
It's Fun to Play the Piano ... Please Pass It On!
|
|
75 members (bluebilly, accordeur, BillS728, aphexdisklavier, bobrunyan, anotherscott, AaronSF, apianostudent, 16 invisible),
2,075
guests, and
354
robots. |
Key:
Admin,
Global Mod,
Mod
|
|
|
Joined: Jan 2009
Posts: 5
Junior Member
|
Junior Member
Joined: Jan 2009
Posts: 5 |
[quote=DrTodd As a computer science Ph.D., maybe I know what I'm talking about. Or not. A tuning algorithm must work by making only 1 adjustment per string to be practical. [/quote] Why would you think that? What does practical even mean in this context? If I update key A's frequency and find a lower entropy and then update key B's frequency and find a yet lower entropy then that is equivalent to updating A and B's frequencies at the same time and find a lower entropy. I've used genetic algorithms successfully in the past for several problems similar to this. The current algorithm is in a sense a very restricted genetic algorithm that doesn't do any cross-breading but only does a single mutation (the keys +/- 1 cent) per generation. It might make sense to do more mutation in the earlier repetitions. That would possible make convergence faster and allow you to hill-climb out of local minima that are far from a global minima. I'm not 100% sure that cross-breading will work...that is why it is called research. If it does work then it would also really help convergence and getting close to a global minima.
|
|
|
|
Joined: Apr 2008
Posts: 2,481
2000 Post Club Member
|
2000 Post Club Member
Joined: Apr 2008
Posts: 2,481 |
[quote=DrTodd As a computer science Ph.D., maybe I know what I'm talking about. Or not. A tuning algorithm must work by making only 1 adjustment per string to be practical. Why would you think that? What does practical even mean in this context? If I update key A's frequency and find a lower entropy and then update key B's frequency and find a yet lower entropy then that is equivalent to updating A and B's frequencies at the same time and find a lower entropy. I've used genetic algorithms successfully in the past for several problems similar to this. The current algorithm is in a sense a very restricted genetic algorithm that doesn't do any cross-breading but only does a single mutation (the keys +/- 1 cent) per generation. It might make sense to do more mutation in the earlier repetitions. That would possible make convergence faster and allow you to hill-climb out of local minima that are far from a global minima. I'm not 100% sure that cross-breading will work...that is why it is called research. If it does work then it would also really help convergence and getting close to a global minima. [/quote] Dr.Todd, I'm a reasonably intelligent person especially in regards to tuning, math, science ect...but the vast majority of what your stating here is totally lost on me...and presumably on 99% of the people here who have no idea about the intricacies and inner workings of algorrithms, entropies, cross-breading, convergences, global/minimal minima ect... Instead of PHD's most of us techs are holders of what can be called DHP's, (doctorate hindered people)...any way to "dumb down" your technical explanation so that we can understand it?
Piano Technician George Brown College /85 Niagara Region
|
|
|
|
Joined: May 2010
Posts: 2,515
2000 Post Club Member
|
2000 Post Club Member
Joined: May 2010
Posts: 2,515 |
Remarkable to me is how, for the first time in any published scientific literature, I am seeing something remarkably close to what aural tuners do. This 'unequal' temperament has been scoffed at, as some kind of human error that ETDs have subsequently improved upon. We can see here in clear scientific demonstration how this work has produced very similar unequal deviation that incredibly good aural tuning produces: the red curve generated by his entropy reduction Monte-Carlo scheme and the black, by a skilled aural tuner. Notice that the computer tuning is sometimes off by a very large amount from the aural tuning, looks like up to 10 cents around key 33. No mention is made in the paper of how the computer tuning actually sounds, so it is not clear to me it actually works. In the past I've recorded each string and computed a tuning optimizing the balance between octaves, fifths, and fourths. The tuning curve looked similar as in this paper, i.e., with a lot of irregularities, but did not sound good at all. The method proposed here requires every single string to be recorded prior to computing the tuning. That doesn't sound practical. I wonder what the result would be if the entropy is calculated from an IH curve fitted to a few measurements, as currently implemented in tunelab for example. If it sounds good it could become practical. I like the concept of the paper. Kees
|
|
|
|
Joined: May 2010
Posts: 2,515
2000 Post Club Member
|
2000 Post Club Member
Joined: May 2010
Posts: 2,515 |
A tuning algorithm must work by making only 1 adjustment per string to be practical.
Why would you think that? What does practical even mean in this context? Sorry I was unclear. I was referring to the requirement to measure each single string of the piano, which is too time consuming to be practical in the field. If I update key A's frequency and find a lower entropy and then update key B's frequency and find a yet lower entropy then that is equivalent to updating A and B's frequencies at the same time and find a lower entropy. I've used genetic algorithms successfully in the past for several problems similar to this. The current algorithm is in a sense a very restricted genetic algorithm that doesn't do any cross-breading but only does a single mutation (the keys +/- 1 cent) per generation. It might make sense to do more mutation in the earlier repetitions. That would possible make convergence faster and allow you to hill-climb out of local minima that are far from a global minima. I'm not 100% sure that cross-breading will work...that is why it is called research. If it does work then it would also really help convergence and getting close to a global minima.
I don't think speed of the optimization algorithm is what needs work, but to determine if the result sounds good. Nevertheless interesting of course. Kees
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
Those kind of algorythm are what relates most to what the tuner is doing.
I can witness of the musicality of a tuning computed using rules in regard of "entropy" (beats) without particular intention as I have seen on most ETD (i.e.n having 1/3 beat at a certain partial level in temperament octave)
This seem to find a set of intervals based on the consonance of the given pianos, at some point it meet what the tuner is doing with the acoustical properties of the instrument and the room.
What could be added is a "level of tempering ratio" or something similar.
How practical to really use ETD are is is a different matter, I find them too slow, despite I am not tuning fast, and only an human can understand how much give have a tuning pin, how to get similar stress on the tuning pin side and on the wire sounding side, etc...
The kind of software is acting as a "beat reduct goodie" in the end.
For my days they are numbered but I stopped counting ....
Last edited by Kamin; 09/27/12 02:53 AM.
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
Of course each note is to be recorded/analyzed it should be each string for excellence, but testing accuracy and the quality of the samples is a good part of the software. Then possibly other effects will arise about final pitch perceived, due to interactions between strings and bridge. I always have perceived tuning as a reconciliation between a tuning scheme or theory, and the real acoustics of the instrument, which relates more or less well to the first. Add a bit of human and you are done
Last edited by Kamin; 09/27/12 02:51 AM.
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
Joined: Jan 2009
Posts: 5
Junior Member
|
Junior Member
Joined: Jan 2009
Posts: 5 |
Why would you think that? What does practical even mean in this context?
If I update key A's frequency and find a lower entropy and then update key B's frequency and find a yet lower entropy then that is equivalent to updating A and B's frequencies at the same time and find a lower entropy. I've used genetic algorithms successfully in the past for several problems similar to this. The current algorithm is in a sense a very restricted genetic algorithm that doesn't do any cross-breading but only does a single mutation (the keys +/- 1 cent) per generation. It might make sense to do more mutation in the earlier repetitions. That would possible make convergence faster and allow you to hill-climb out of local minima that are far from a global minima. I'm not 100% sure that cross-breading will work...that is why it is called research. If it does work then it would also really help convergence and getting close to a global minima.
Dr.Todd, I'm a reasonably intelligent person especially in regards to tuning, math, science ect...but the vast majority of what your stating here is totally lost on me...and presumably on 99% of the people here who have no idea about the intricacies and inner workings of algorithms, entropies, cross-breading, convergences, global/minimal minima ect... Instead of PHD's most of us techs are holders of what can be called DHP's, (doctorate hindered people)...any way to "dumb down" your technical explanation so that we can understand it? Imagine a vast mountain range. This corresponds to all possible tunings. The peaks are bad tunings, the valleys are good tunings. The goal is to find the valley with the lowest elevation (the best tuning). In the original paper, the approach is to drop one guy off at a very specific point in the mountain range corresponding to pure equal temperament. He then looks around him a short distance and chooses to walk a direction that seems to go down. He repeats this process until he can't go down any more. This process is sensitive/unstable for imagine that he started on a peak where left and right both go down. Left may lead to a much lower valley but he can't see that and so in randomly picking right he has limited how low he can go. In a genetic algorithm, you drop thousands of guys in the mountain range at random with GPS altimeters and radios. Some of the guys that are lower than the others do similar to the first guy...they pick some down direction and start walking. The guys that are higher than the others pick two of the guys (at random) that are lower than them and walk to some midpoint between them. The two guys may be far apart and so there may be a huge peak in the middle. Or you could get lucky and there be a valley. Most of the time it doesn't improve the situation but with hundreds of guys doing it usually someone will find some lower elevation. Repeat until nobody manages to find a lower spot for a while. You can see in the first case that there may be much lower valleys in the vast range than the one guy managed to find. In the latter case, there's no guarantee of finding the absolute lowest valley but your chances are higher and the lowest valley you find is likely to be much lower than with just one guy. Because they have radios, they can quickly shift focus to promising areas (this is convergence speed).
|
|
|
|
Joined: Jan 2009
Posts: 5
Junior Member
|
Junior Member
Joined: Jan 2009
Posts: 5 |
Remarkable to me is how, for the first time in any published scientific literature, I am seeing something remarkably close to what aural tuners do. This 'unequal' temperament has been scoffed at, as some kind of human error that ETDs have subsequently improved upon. We can see here in clear scientific demonstration how this work has produced very similar unequal deviation that incredibly good aural tuning produces: the red curve generated by his entropy reduction Monte-Carlo scheme and the black, by a skilled aural tuner. Notice that the computer tuning is sometimes off by a very large amount from the aural tuning, looks like up to 10 cents around key 33. No mention is made in the paper of how the computer tuning actually sounds, so it is not clear to me it actually works. In the past I've recorded each string and computed a tuning optimizing the balance between octaves, fifths, and fourths. The tuning curve looked similar as in this paper, i.e., with a lot of irregularities, but did not sound good at all. The method proposed here requires every single string to be recorded prior to computing the tuning. That doesn't sound practical. I wonder what the result would be if the entropy is calculated from an IH curve fitted to a few measurements, as currently implemented in tunelab for example. If it sounds good it could become practical. In the paper, he mentions that they tried looking at intervals like piano tuners do rather than overall entropy and that this destabilized the method and drove the whole system out of equal temperament closer to just intonation. For me, it took maybe 45 minutes to make the recordings. Dirk's piano tuner does something similar but has an integrated recorder which makes the process faster...probably down to around 20 minutes. You only have to do this once per piano so repeated tunings would be a lot faster. It depends who would use this work though. It might not speed things up much for a professional tuner but for a hobbyist you only have to do it once and so it doesn't matter.
|
|
|
|
Joined: Feb 2011
Posts: 4,331
4000 Post Club Member
|
4000 Post Club Member
Joined: Feb 2011
Posts: 4,331 |
In the paper, he mentions that they tried looking at intervals like piano tuners do rather than overall entropy and that this destabilized the method and drove the whole system out of equal temperament closer to just intonation. What is exciting about the paper is that matching up all the partials and the piano's other tones using communication theory produced results very similar to aural tuning in equal temperament. The observation about intervals and just intonation indicates that changing the criteria may result in other temperaments. I wonder if different weights applied to the intervals associated with the keys round the circle of fifths would produce recognizable well temperaments. The paper mentioned there are many local minima. It would be in interesting to see the spread of your genetic solutions. Keep up the good work and let us know the results - especially comparisons with aural tuning.
Ian Russell Schiedmayer & Soehne, 1925 Model 14, 140cm Ibach, 1905 F-IV, 235cm
|
|
|
|
Joined: Apr 2010
Posts: 2,326
2000 Post Club Member
|
2000 Post Club Member
Joined: Apr 2010
Posts: 2,326 |
. You only have to do this once per piano so repeated tunings would be a lot faster. It depends who would use this work though. This is an assumption made by machine-centric designers before encountering the real world. In actuality, machine-tunings on fine instruments do change over time. That is to say, a machine doing a tuning from scratch may come up with a different tuning on a different date than the saved tuning for the same instrument.
Keith Akins, RPT Piano Technologist USA Distributor for Isaac Cadenza hammers and Profundo Bass Strings Supporting Piano Owners D-I-Y piano tuning and repair editor emeritus of Piano Technicians Journal
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
I also can guarantee you that machines have no soul, while I seem to recall a movie where the opposite was stated, just a movie, probably... ETD my darling, please come home
Last edited by Kamin; 09/28/12 12:23 AM.
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
Joined: Oct 2006
Posts: 2,677
2000 Post Club Member
|
2000 Post Club Member
Joined: Oct 2006
Posts: 2,677 |
ETDs ('sorry ETD guys) are slow and cumbersome. They last maybe ten years on average due to equipment entropy and software obsolescence. Humans last an average of 80 years and can tune effectively for 50, or more, of those years.
In terms of numbers ... HUMANS win!
(I know that wasn't the thrust of the OPs question, but, to me, it's an interesting twist on an answer.)
|
|
|
|
Joined: Apr 2008
Posts: 2,481
2000 Post Club Member
|
2000 Post Club Member
Joined: Apr 2008
Posts: 2,481 |
ETDs ('sorry ETD guys) are slow and cumbersome. They last maybe ten years on average due to equipment entropy and software obsolescence. Humans last an average of 80 years and can tune effectively for 50, or more, of those years.
In terms of numbers ... HUMANS win!
(I know that wasn't the thrust of the OPs question, but, to me, it's an interesting twist on an answer.) ETD's can be utilized efficiently with "auto detect" or foot operated mouse clicker to move from note to note with no appreciatable time loss. The response time to the visual display from the brains reception of it through your optic nerves is actually slightly faster than the accoustic alternative using your ear. Read up a bit on the effects of aging on the human ear, both from the perspective of hearing loss and also from perspective of frequency judgement. The latter refers to the "taste aspect" of choosing appropriate artificial stretch. As we age we begin to regard higher frequencies normally found in the upper treble as being flatter than what they actually are. Machines will not change in this respect. Microphones also do not lose their ability to detect sounds as they age the way our ears do. Nor do they get annoyed or frustrated when other noises overbear the sounds it hears...it might flicker on the display but a decent reading usually comes through. Software based ETD's can be transferred to newer machines quite easily as the machines age. The better ETD's in good hands that know how to fully utilize them now will produce results equivilent to fine aural tunings 95-100% of the time. (supported with the tune off experiments of past). When the effectiveness is this high to begin with, obsolescence becomes a moot point. There is an arguement to be made that a human who has tuned for 50 years, has ingrained their process so deeply that should improvement be needed, they are more resistant to change than an ETD will be with a software upgrade. Another of an ETD over aural tuning is that your reference pitch A440 is actually A440 to the full accuracy of the machines resolution/repeatability (if you can stabilize the spinner). Not so with a tuning fork, which can vary with temperature or the forks condition, compounded by the inability of the human ear to be as accurate in discerning an extremely slow roll inaccuracy on the setting which an ETD spinner will easily display. As for Kamins' comment about machines not having a soul....not all souls are decent or appropriate for things. Charles Manson has a soul, but I would sooner use a robot than him if they had to be around other people.
Last edited by Emmery; 09/28/12 10:05 AM.
Piano Technician George Brown College /85 Niagara Region
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
please dont get upset Emmery. ii am joking!
but what you state about ear vs eye reaction is false ( at last for piano tuning, but the ear induce a more immediate reflex than the eyes, to me)..
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
Joined: Apr 2008
Posts: 2,481
2000 Post Club Member
|
2000 Post Club Member
Joined: Apr 2008
Posts: 2,481 |
kamin, actual time for nerve signals travelling to the brain optically is faster than accoustically. The latter process relies on physical attributes of the hairs in the cochlea to be stimulated where visually it is instantaneous (or at the speed of light). A bright flash will startle you much faster than a loud bang.
More importantly, for piano tuning, take an extremely weak partial which has an unusually short duration for you to percieve it fully. Because you are using comparison and beat rates for accoustic setting, a short duration will negate your effectiveness at rendering a slow beating interval. If the duration is less than the beat rate itself in cps, you cannot accoustically set it at all. That same short duration of on an ETD reading of the same partial can still visibly convey more information (a slight movement on the spinner) than you will hear with your ears...an entire cycle on the spinner is not needed, even a fraction of the full cycle will indicate sharp or flat.
(added) An additional benefit of the ETD is that it reads each note individually. Take for example an unstable string which will be slightly sharp on attack, stabilize for a fraction of a second, and then slowely roll flat for the remainder of the sustain. If one is using two notes (by ear) to tune, the best you can do is percieve the entire sound to the lowest common denominator of the two (as far as sustain goes). When either of the two notes dies out sooner, you are finished as far as your perception of beat speeds goes. Visual recognition takes each note for itself and can play an important part in setting an unstable string to a later decay speed (if that decay frequency dominates the overall tone).
Last edited by Emmery; 09/28/12 11:27 AM.
Piano Technician George Brown College /85 Niagara Region
|
|
|
|
Joined: Jun 2003
Posts: 32,060
Yikes! 10000 Post Club Member
|
Yikes! 10000 Post Club Member
Joined: Jun 2003
Posts: 32,060 |
There was just a bill signed in California allowing self-driving cars. The next step is making them so they can accept any fuel. Then you will get into a car, tell it where to go, and be deposited there as automobile droppings once our mechanical masters attain dominance. So yes, the days of human piano tuners are numbered! As are everyone's!
Semipro Tech
|
|
|
|
Joined: Apr 2008
Posts: 2,481
2000 Post Club Member
|
2000 Post Club Member
Joined: Apr 2008
Posts: 2,481 |
There was just a bill signed in California allowing self-driving cars. The next step is making them so they can accept any fuel. Then you will get into a car, tell it where to go, and be deposited there as automobile droppings once our mechanical masters attain dominance. So yes, the days of human piano tuners are numbered! As are everyone's! this does not surprise me since there are self parking cars out there already. Now if a well imbibed Scotsman could be understood by its navigation equipment and actually delivered to where he wanted to go...well that would be impressive.
Piano Technician George Brown College /85 Niagara Region
|
|
|
|
Joined: May 2012
Posts: 7,439
7000 Post Club Member
|
7000 Post Club Member
Joined: May 2012
Posts: 7,439 |
actual time for nerve signals travelling to the brain optically is faster than accoustically. The latter process relies on physical attributes of the hairs in the cochlea to be stimulated where visually it is instantaneous (or at the speed of light). A bright flash will startle you much faster than a loud bang.
Well, not exactly. The difference between the speed of sound and the speed of light is not the issue as the speed of sound, traveling from the soundboard to the ear or microphone, is virtually the same since the distance would be so similar. Both are *listening* for sonic transmission from the same source. It becomes a subsequent response to look at the ETD screen. Within the body, the conversion of optic or sonic stimulus, into electricity borne by salt ions for neural transmission across the synapses, is equivalent. Electrical transmission through a metal is conducted by means of electrons. Neural transmission is by means of ions. The flash-bang delay of fireworks doesn't apply. For more information, and a very interesting read, I recommend: "The Spark of Life - Electricity in the Human Body" by Frances Ashcroft.
Marty in Minnesota
It's much easier to bash a Steinway than it is to play one.
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
that is how I perceive it. if we need so much time to push on the brake pedal when we see a danger, I suppose that if we where driving only with audible signals we would push on the pedal way sooner. I seem to recall an article stting a 1 for 6 or even larger ratio.
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
Joined: Mar 2008
Posts: 9,230
9000 Post Club Member
|
9000 Post Club Member
Joined: Mar 2008
Posts: 9,230 |
24 images second, definition semm really crude when we think of it
Professional of the profession. Foo Foo specialist I wish to add some kind and sensitive phrase but nothing comes to mind.!
|
|
|
|
|
|
Piano
by Gino2 - 04/17/24 02:34 PM
|
Piano
by Gino2 - 04/17/24 02:23 PM
|
|
Forums43
Topics223,408
Posts3,349,457
Members111,637
|
Most Online15,252 Mar 21st, 2010
|
|
|
|
|
|