Can Old Italian sound be seeen in violin body impact spectra?

[Anders Buen]

Heinrich Dünnwald studied sine swept violin spectra from some 700 instruments in his PhD work from the late 80ties and early 90ties. He measured spectra from some 55 old Italians, 75 old master instruments made before 1800, 300 master instruments made after 1800, 170 factory made violins and 100 of another category (probably amateur made instruments etc). He measured all instruments using the same method exciting the instruments by a very light exciter pushing against the bridge G string side giving a close to flat force input. All instruments were measured the same way in an echo and noise free environment. (His excitation signal was very weak, a penalty for the lightness of the excitation system and keeping a small influence on the violin response)

His measurement thechnique would be equivalent to using an impact hammer system where the input force signal is accounted for.

One of the main findings was that the level of the A0 resonance compared to the mid frequency range from 650-1120Hz tended to be stronger in the old Italians as compared to most of the rest of the instruments.

I enclose two figures illustrating that point. One show how the loudness of the A0 was measured out of the violin spectra in his study and the other figure show plots of the cumulative distribution of that relative A0 level from all the 700 instrument spectra sorted on these groups of instruments:

a ) Factory violins

b ) Master violins made after 1800

c ) Master violins made before 1800

d ) Old Italians

He chose an arbitrary relative A0 level of 18 dB as being the requirement for Old Italian sound trait of that category. That translates to a 7dB level difference between the A0 and the stronges resonance in the region from 650-1120Hz in the spectra.

From the cumulative distribution you see that about 90 % of the Old Italians have relative levels (L) over 18dB. Among the master violins from before 1800 there are about 45% of the instruments meeting the criterium and for the master instruments after 1800 only 25% does. Only about 15% of the factory violins meet that A0 level criterion.

Dünnwald also had some other factors he extracted from the frequency spectra dividing better sounding instruments from the lesser sounding ones. I may return to that later. But the relative level of the A0 was the main factor dividing the instrument groups. Only 34 % of the instruments in the study passed that "quality parameter". The other factors made the group of "very good violins" even smaller.

One could discuss the significance of this finding, but I just give the principles yet. And then I will post some preliminary results as to how one can increase the probability of making an instrument meeting and exceeding Dünnwalds main spectrum indicator for "Old Italian sound", the relative level of the A0

[Anders Buen]

About a year now I have had an impact hammer rig up and working here and during my last summer holiday I started to systematize measurements from that rig into one of my data mining sets I have been working on for a few years now. I have some 31 data entries from that impact hammer rig now with data from about equal number of violins and hardangers. That is not an impressing number comparing to Dünnwalds 700 from his three - four year PhD work, but it is not bad taking into account that I do this besides my regular full time job, and my hardanger making and repair business is a very small one. It becomes even smaller from this "geek project". :-) But the making and repair business give access to instruments to entry data from. There are data from a few very good to mediocre instruments, some master built and some recent chinese violins ranging from about 1860 up to now. No old italians yet. :-)

Along with the recorded impact hammer spectra also a quite huge set of data from the instruments are recorded like thicknesses, arching heights, dimensions, wood sound speed data, wood density, plate weights if tops come off, different tap tones, f-hole lenghs, distances bewteen them, bridge properties, sound post position, etc. Behind each "data entry" of an instrument there is about 200 data inputs with a main weight on the recorded graduations.

Now back to the topic: I enclose my cumulative curve for the L parameter (relative level of the A0) from my data set. It is sorted on hardangers and violins as different groups. Hardangers are black, violins are red. The data set is still small. But you may see a tendency that the hardangers tend to have a little stronger A0 levels as compared to the violins in the group. About 60% of the hardangers and about 50% of the violins in the set meet Dünnwalds criterion of having L >= 18 dB.

Now is it possible to find any information in the data set that correlate with the variation in the measured values of that A0 relative level?

[Johnmasters]

About a year now I have had an impact hammer rig up and working here and during my last summer holiday I started to systematize measurements from that rig into one of my data mining sets I have been working on for a few year now. I have some 31 data entries from that impact hammer rig now with data from about equal number of violins and hardangers. That is not an impressing number comparing to Dünnwalds 700 from his three - four year PhD work, but it is not bad taking into account that I do this besides my regular full time job, and my hardanger making and repair business is a very small one. It becomes even smaller from this "geek project". :-) But the making and repair business give acess to instruments to entry data from. There are data from a few very good to mediocre instruments, some master built and some recent chinese violins ranging from about 1860 up to now. No old italians yet. :-)

Along with the recorded impact hammer spectra also a quite huge set of data from the instruments are recorded like thicknesses, arching heights, dimensions, wood sound speed data, wood density, plate weights if tops come off, different tap tones, f-hole lenghs, distances bewteen them, bridge properties, sound post position, etc. Behind each "data entry of an instrument there is about 200 data inputs with a main weight on the recorded graduations.

Now back to the topic: I enclose my cumulative curve for the L parameter (relative level of the A0) from my data set. It is sorted on hardangers and violins as different groups. Hardangers are black, violins are red. The data set is still small. But you may see a tendency that the hardangers tend to have a little stronger A0 levels as compared to the violins in the group. About 60% of the hardangers and about 50% of the violins in the set meet Dünnwalds criterion of having L >= 18 dB.

Now is it possible to find any information in the data set that correlate with the variation in the measured values of that A0 relative level?

Do I misread the graph? It seems to say delta-L less than 7dB in old Italians. Can you say anything about the damping of A-naught? Can you see a total area under the peak.. I am thinking that large motions of a f-hole wing might spread out the peak of the A-naught. At least that would be a large motion which could have greater or lesser hysteresis damping.

[Anders Buen]

Do I misread the graph? It seems to say delta-L less than 7dB in old Italians. Can you say anything about the damping of A-naught? Can you see a total area under the peak.. I am thinking that large motions of a f-hole wing might spread out the peak of the A-naught. At least that would be a large motion which could have greater or lesser hysteresis damping.

Delta L =< 7 dB in the spectrum corresponds to Dünnwalds L >= 18 dB because he adjusted the strongest resonance in the 650-1120Hz region to be at 25dB.

25-7 dB = 18 dB :-)

[Anders Buen]

For some time I have been using excel spreadsheets for the data analysis. Ok, but a bit cumbersome to work with. Recently I have started to learn a statistics software package, hopefullly to improve a bit on the statistics, and have some fun.

One way to present such data is to use scatterplots for each "input" parameter compared to e.g. the L. You have seen I have posted some of these in earlier posts. Its an effective way to look for pattterns in data and how they may correlate. But many facors may influence. How do they e.g. interact?

I have tried some multiple regression, e.g. on Dünnwalds L parameter (relative loudness of the A0) and one model that could "explain" about 55% of the variation in the L parameter data in my set was one containing the top plate weight and the height of the bassbar at the widest part of the upper bout. I measure the height of the bassbar from the inside of the plate.

I will not post scatterplots here now, nor the regression formula. But the statistics package has an option to make a color surface plot of the model. I do not know exactly how this is made but it is probably an interpolation algorithm working behind it. The raw data would look more like a cloud of dots. But the figure give an intuitive idea of how the top plate weight and bassbar upper height may interact and influence the relative level of the A0.

I do not think that there is a guarantee to get a L-parameter >= 18 dB by using this figure, but you will proably increase the proablility for getting that. Does this comply with workshop intuition?

(All this information just for posting a little "statistics based artwork", puh! :-) )

[Magnus Nedregard]

Hi Anders, for the simple minded non-acoustics among us, allow me to introduce a few really silly questions: Could we say something about how high levels of the AO relates to the mechanical function of the violin. I guess the AO is facilitated by a flexible corpus shell, as that must be what puts it into action after all, but as we know; to make a thin-shelled violin is basically not the solution, although I guess it would typically have

responsive air modes?

But could the presence of cracks, patches and worn edges be responsible for some acoustical differences. I sometimes close open seams on an instrument, and actually; tonally the effect is not necessarily an improvement. How would a certain number of more or less open cracks, (as is the situation on quite a few classical instruments, even the ones in "good condition") affect sound? Typically a nice old fiddle has a very sturdy center, with lots of patches and cleats in the f-hole vicinity, combined with loose f-wings, and more often than not little hairline cracks close to the neck and button blocks. Would such factors make a typical, acoustically recognizable effect?

[David Burgess]

I wonder if strength at A0 is itself important, or whether it is an artifact of something else which is important.

When I experimented with a graphic equalizer to push the levels around at about 270 hz, sound quality was rather unresponsive compared to changing the levels at other frequencies.

[Anders Buen]

Hi Anders, for the simple minded non-acoustics among us, allow me to introduce a few really silly questions: Could we say something about how high levels of the AO relates to the mechanical function of the violin. I guess the AO is facilitated by a flexible corpus shell, as that must be what puts it into action after all, but as we know; to make a thin-shelled violin is basically not the solution, although I guess it would typically have responsive air modes?

I generally think your intuition is correct. A more flexible body will make the A0 stronger in relation to the mid frequency region, at least it will make the A0 stronger. Now there probably is an upper limit here. I would guess related to hollowness. There is of course a structural issue also, leading to your following sentence.

But could the presence of cracks, patches and worn edges be responsible for some acoustical differences.

That could possibly be. I do not come across a lot of these old valuable instruments, but I do glue in patches in back plates. My experience with pateched tops is very limited. But my first guess would be that these patched up instruments will move in the direction of a weaker A0 and thus L parameter.

If you bring some some intruments of that category for measurements, we may get better answers to that question.

I sometimes close open seams on an instrument, and actually; tonally the effect is not necessarily an improvement. How would a certain number of more or less open cracks, (as is the situation on quite a few classical instruments, even the ones in "good condition") affect sound? Typically a nice old fiddle has a very sturdy center, with lots of patches and cleats in the f-hole vicinity, combined with loose f-wings, and more often than not little hairline cracks close to the neck and button blocks. Would such factors make a typical, acoustically recognizable effect?

I do not know for sure. I have seen effects of gluing cracks and getting a better bond for the top on the ribs. The quality of the glue seam does seem to matter for the stiffness of the body and seem to affect sound and playability. But again, my experience is limited. It may depend where the cracks are as well. A crack south of the f-holes seems important, a f-hole outer wing crack: less so.

Again if you come across such intruments I would be glad to include them in the database and we may draw concusions from the possible effects.. :-)

I think the studiness of the central part of the top is important. Do you think that these instrumetns with patches has a more sturdy central portion than most mass produced violins?

[JimMurphy]

Interesting stuff, Anders.

So, what's the next step - to compare only A0's of Old Italians sounding smooth & clear, i.e., drop the noisy ones ??

Jim

[Anders Buen]

I wonder if strength at A0 is itself important, or whether it is an artifact of something else which is important.

When I experimented with a graphic equalizer to push the levels around at about 270 hz, sound quality was rather unresponsive compared to changing the levels at other frequencies.

I think that it may be both. The complete region between the A0 and the B1- is affected by the same thing that may affect the strength of the A0. That is: the played notes in almost one and a half octave width may be affected. The D string is generally rich sounding if the L parameter is high, at least that is my impression.

But it is not just the A0, the L parameter take into account what happens in the mid frequency range too.

I would like to measure the stiffness of the plate directly. I think that may give some useful info. But I need to do an investment first.

I think the type of music played is important to what effect one get from adjusting the strength of the A0-B1- region, especially for the effect on the A0. One of the played notes has to be pretty close to the A0 in order to hear the effect clearly.

The A0 is driven for any played note by the starting transient as we see in the post about the Ole Bull del Gesu and Miracle Makers instruments here: http://www.maestronet.com/forum/index.php?...st&p=444401

[Don Noon]

Anders,

Great stuff. First, did the Dunnwald tests have a chinrest?

I agree with David, in that I don't see this L parameter saying something important about A0. It looks to me like this L parameter is more a measure of the strength of the response around 1000 Hz, and just happens to be measured relative to A0. I would rather see an "average energy level" for the entire spectrum used as a baseline, and then measure the height of the Dunnwald bands relative to that.

The features of the spectra that stand out as "old Italian" is, IMHO, the relatively low response around 1000 Hz and the strong "bridge hill" between 2000 and 3000 Hz. In good modern instruments, there is a much stronger response around 1000 Hz, and the "bridge hill" appears at a higher frequency, 4000 Hz or so. In crappy instruments, there is a huge relative strength around 1000 Hz, and they're dead at the high end.

I have taken impact spectra of a few old Italian fiddles, 2 Strads, 1 Guarneri, an Amati, and a few others like Carcassi. Not all of them show the "depressed 1000 Hz, strong 2000 - 3000 Hz" feature, but the 2 Strads definitely did. And, in general, it seems as if the 1000 Hz region is on average quite a bit lower than everything else. And I haven't measured anything yet that comes close to looking like the spectrum of the two Strads. Interestingly, the Titian response looks more like a modern instrument to me, with quite a strong peak around 1000 Hz.

At the moment, I can not come up with a clear explanation of how to obtain a response that is relatively low around 1000 Hz, but still strong elsewhere. I will just speculate that since the archings and graduations are not anything extraordinary, and certainly have been copied extensively without(?) getting the same results, therefore the properties of the wood seem to be the leading candidate. Joseph Curtin's article hinted of a possibility of stiffness ratio abnormalities (high stiffness along grain, weak across grain) in old Italian instruments, and I think there could well be damping properties involved too. All speculation on my part.

[Johnmasters]

Delta L >= 7 dB in the spectrum corresponds to Dünnwalds L >= 18 dB because he adjusted the strongest resonance in the 650-1120Hz region to be at 25dB.

25-7 dB = 18 dB :-)

The graph has < and not >. So I am not seeing the proper normalization.

But could you answer the more important point, about the possible total damping in the A-naught mode.

[Wm. Johnston]

Heirich Dünnwald studied sine swept violin spectra...

But in the second picture you wrote impact spectra. Which method was used on this plot?

[Wm. Johnston]

Now back to the topic: I enclose my cumulative curve for the L parameter (relative level of the A0) from my data set. It is sorted on hardangers and violins as different groups. Hardangers are black, violins are red. The data set is still small. But you may see a tendency that the hardangers tend to have a little stronger A0 levels as compared to the violins in the group. About 60% of the hardangers and about 50% of the violins in the set meet Dünnwalds criterion of having L >= 18 dB.

Now is it possible to find any information in the data set that correlate with the variation in the measured values of that A0 relative level?

CDF's all pretty much look the same to me. I'm curious what the PDF looks like, it's usually easier to see how things are distributed then. How close to a Gaussian (Normal) is the distribution?

Seems to me that if this parameter were really indicative of how good a violin sounds then if you had a random collection of violins and also a set of good sounding ones then the random collection would have a Gaussian distribution but the 'good' violins would cause the distribution to be skewed one way or be bimodal.

[Anders Buen]

The graph has < and not >. So I am not seeing the proper normalization.

Yes, sorry for the incorrect inequality direction.

But could you answer the more important point, about the possible total damping in the A-naught mode.

I do not measure damping so the answer would simply be a specualtion. The speculation is that I do not think that f-hole wing flapping is important for the A0 damping.

[Anders Buen]

But in the second picture you wrote impact spectra. Which method was used on this plot?

Sine swept excitaiton via the transducer he developed for the purpose. A highly damped string carrying current in a static magnetic field. I used the phrase "impact hammer spectrum" as that is what I think we are most used to here.

[Anders Buen]

CDF's all pretty much look the same to me. I'm curious what the PDF looks like, it's usually easier to see how things are distributed then. How close to a Gaussian (Normal) is the distribution?

Seems to me that if this parameter were really indicative of how good a violin sounds then if you had a random collection of violins and also a set of good sounding ones then the random collection would have a Gaussian distribution but the 'good' violins would cause the distribution to be skewed one way or be bimodal.

Maybe you can read out that information from this figure? Not perfect, but not really bad either. Early days yet.

[Anders Buen]

Interesting stuff, Anders.

So, what's the next step - to compare only A0's of Old Italians sounding smooth & clear, i.e., drop the noisy ones ??

Jim

Unfortunately there is no good information in Dünnwalds work on how the insturments sound, except for 'nasality', 'harshness' and 'clarity'. I would believe harshness is somewhat related to 'noisy' instruments.

He does not revel how the instruments were judged. Therefore Claudia Fritz and the 'Cambridge group' with Jim Woodhouse, Ian Cross and Brian Moore has been working on that subject. How relevant are the frequency bands he used and the subjective notations in relation to combinations of these? There will come some publications in the Journal of the Acoustical Society of America (JASA) on the subject in the near future.

One of their findings is that 'nasality' seem to be interpreted in at least two different ways by different listeners.

But no matter what they may find, Dünnwalds statistics does distinguish between instrument groups and sort out most of the Old Italians. At least that is what he claim. He denote them as being 'very good sounding' but without a reference to how they have been judged, as far as I understand. We may speculate that they are instruments in use by professional musicians.

[Johnmasters]

I do not measure damping so the answer would simply be a specualtion. The speculation is that I do not think that f-hole wing flapping is important for the A0 damping.

It was something that I thought might fix the reference level at the peak of the A-naught. I can see that one might need a lot of damping to change the peak height enough to say anything. Just a thought.

And I suppose it would go in the wrong direction to provide any explanation. I probably would not have suggested it if I had known that the inequality was an actual error.

[Oded Kishony]

Don Noon writes:

At the moment, I can not come up with a clear explanation of how to obtain a response that is relatively low around 1000 Hz, but still strong elsewhere

One way to decrease radiation at a given frequency is to increase phase cancelling.

Question: If a frequency at ~500Hz were very strong would it have the effect of diminishing the harmonic of that frequency (100Hz) ?

Oded

[Anders Buen]

It was something that I thought might fix the reference level at the peak of the A-naught. I can see that one might need a lot of damping to change the peak height enough to say anything. Just a thought.

And I suppose it would go in the wrong direction to provide any explanation. I probably would not have suggested it if I had known that the inequality was an actual error.

I did measure A0 and signature mode damping in violins in my master thesis. The A0 tend to be more damped than the body structural modes, about double. However many modes are a mixture of body and air movement in the f holes. If the B0 fingerboard neck bending mode is tuned to be close, or just is close to the A0, then the damping of the A0 alone will be difficult to measure. The mode may look wider than it is if left alone to vibrate.

Bissinger found that the B1- and B1+ had a substantial degree of its sound radiation from the f-holes. The more sound radiation, the more damping.. But we hold the instruments for playing, adding damping to an extent that will dominate anything else, but maybe not the A0?

The inner f-hole wings do seem to intereact in the "body hill cluster", at least that was Jansson and Dunlops finding. I may return to the effect the lenght of the f holes, and thus their wings, may have on the 'noise' high frequency region. This is one of the differences we may see between violin and hardanger instrument spectra as hardangers generally have longer f-holes and possibly a weaker "inverse tube effect" from a flatter central cross arching between the f-holes.

[Anders Buen]

Anders,

Great stuff. First, did the Dunnwald tests have a chinrest?

Thanks. I do not know if they did have a chinrest or not. Do you think that different chinrests do have an effect on the A0 or the L parameter?

I do have some spectra of both, but havent included them in the dataset. But the statistics package give an opportunity to separate the categories e.g. with and wo chinrests easily. But it's some work to include the pairs of data. Might be one of the next steps.

[Anders Buen]

I agree with David, in that I don't see this L parameter saying something important about A0. It looks to me like this L parameter is more a measure of the strength of the response around 1000 Hz, and just happens to be measured relative to A0. I would rather see an "average energy level" for the entire spectrum used as a baseline, and then measure the height of the Dunnwald bands relative to that.

You may be right, at least there is more of a pattern in the scatterplot bewteen L and the maximum level in the 650Hz-1,12kHz frequency region than with the level of the A0.

The L parameter may also be a sorting between the violins with ligther and heavier tops and/or softer bassbars in the upper bout. There are also just under 50% of the variation that is not accounted for as well in this data set.

One of the problems with Dunnwalds work is that he does not reveal the absolute levels of the spectra. He adjusted the excitation force so that the highest peak touced 25 dB. I think one of the problems was that he had a too weak excitation signal and system. You see that he does not have data e.g. between the A0 and B1- in many of the spectra. He seem to have had problems with his signal strenght. He is looking at relaative levels between the different spectrum parts without reference to the absolute level. I think that gtive a correct interpretition of how timbre works. But loudness is very important too. Dünnwald may have that information in some form, but does not reveal it in his articles.

I do have both in my measurements.

[Anders Buen]

Interestingly, the Titian response looks more like a modern instrument to me, with quite a strong peak around 1000 Hz.

Yes I agree with you in that observation. It also has an exeptionally even response in comparison to many other great instruments.

[Johnmasters]

Thanks, Andres. I know that traditional fiddlers sometimes adjust things by carving their f-holes. It seems an intuitive thing to do because "they are where the sound comes from." Maybe old-time makers adjusted them and got good results even if their mental picture was not correct.