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

[Anders Buen]

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.

Just because you address that theme, I looked for what candidates there are in my set that may influence the B region (650Hz - 1.3kHz). There is no factor seeming to be significant by the strict rule I have for that. But by categorizing the hardangers and violins alone, the instruments with a lower density in the top plates tend to have a weaker response in that frequency region.

However, I do think that the regions next to the B on each side seem to have clearer information on factors that may influence. The A region (190-650Hz) is suppressed by a higher arching. And the D region (1,64Hz-2,58Hz) seem to be increased by high tuned bridges and longer bodies.

You may see the traces of something now that surprised me a bit, that the model (and wood properties) seem to be the most significant factors determining the Dünnwald bands and parameters beyond the L. The bridge and the f-hole length are other factors.

There is a weak trend in my data that instruments with high L values tend to be less loud.

[JimMurphy]

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

Thanks Anders. I'm thinking both nasal & harsh violins [Dünnwald's comments] may be muddying more useful A0-level & frequency band correlations [within his data]. Kinda like when there's a need to examine A440Hz tuning fork "precision" and the sample set includes 435Hz & 445Hz tuning forks.

Jim

[Anders Buen]

I'm thinking both nasal & harsh violins [Dünnwald's comments] may be muddying more useful A0-level & frequency band correlations [within his data]. Kinda like when there's a need to examine A440Hz tuning fork "precision" and the sample set includes 435Hz & 445Hz tuning forks.

Could be, I think you might be talking about masking effects. In general I think the masking effect is stronger upwards than downwards in frequency. I do not know exactly how masking may affect the perception of violin sound, but we may assume that it does in some form, especially for the players.

[Anders Buen]

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.

I have tried that too. It can be effectful, but found that it generally is better to keep the A0 low in frequency for a good low end. There will be many combining effects here. Moving the A0 closer to the B1 modes may affect the radiation, f-hole size alone might also play a role for the radiation alone. At some time we might have an 'acoustic camera' available to compare instruments and how the f-holes radiate..

[Anders Buen]

A 1708 Strad spectrum measured by Dünnwald as an example of one with a high L value, some 24dB or so. I have also read out the frequencies at one time.

[Anders Buen]

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.

There is a also a correlation with the back plate arching height in the set, the higher the back plate arching the weaker the 650-1,3kHz (B ) region. But again the correlation is not significant by the strict rule for that. May return to the question again when there is much more data available.

When it comes to Curtins idea of along grain to across grain stiffness, that can e.g. be measured with compression sound speeds along and across the plates. I do have data for that. This far the sound speed ratios for the back plates seem to give more patterns in the data than what the top show. Cross grain to alonjg grain stiffness ratio effect is highly dependant on the grain angle. Just a few degrees off and the cross grain stiffness drop dramatically. So if that stiffness ratio is a factor that might play a role for Old Italian sound, we would find a higher degree of non parallell wood grain in these instruments compared to e.g. most modern ones. Is that the case?

I think Joe has seen that it is possible to get similar mode 5 to mode 2 frequency ratios in new tops even with the grain parallell to the centre glue line as the old italian tops if the end regions are not made using the stiffening "Sacconi plateus".

[Don Noon]

Anders, do you ever sleep?

Have you found anything that DOES correlate with grain stiffness ratios?

[Anders Buen]

Anders, do you ever sleep?

Have you found anything that DOES correlate with grain stiffness ratios?

Woke up early today as well.. :-)

Yes, especially for the back plate. E.g level of the B1+ seem to correlate with high ratios for the back plate. But the data spread is not good for the statistics here. There is an extreme data point there levereing the data somewhat, but still I think there is a clear trend among the hardangers here. It is not uncommon with softwoods in the back plates of hardangers and the extreme point is from a slab cut black alder back I think.

You may also see that some data points have the same x-value but different y axis values. These are from the same instrument entered twice or in some cases more times with slight alterations e.g. to bridge setup etc, or 'before-after' comparisons. Entering the same data several times also introduces a sort of leveraging of the data that can cause problems. Larger number of data hopefully will 'water out' that problem.

This seem to be an example where the violins and hardangers may behave different of some reson, if its possible to draw a conclucion yet. If that turns out to be the case, I would need to split the data if I would like to get better regression results and models. But then the penalty is less data to mine on and I would need to wait longer to get reliable results.

[Anders Buen]

Have you found anything that DOES correlate with grain stiffness ratios?

Here is another example of scatterplots with sound speed ratios for back plates against summed Dünnwald parameters and one with the former + the overall level.

I will have to explain what the rest of the Dünnwald paramters are now for clarity. That is next, but for short the sums in the plots here are:

L (relative level of the A0 to the 650Hz-1.12kHz region)+ ACD-B + DE-F (first plot)

The above sum + SPL/input Force (second plot)

I have made up these combined parameters myself, as a sort of way to rate the instruments by spectral data. The higher numbers the 'better'. I tend to like the instruments with high values better, but I am not quite sure if that is because I like the idea or if that correlation is real and would emerge if blind tests were conducted. I haven't been rating the instruments on subjective evaluations sheets systematically. That may be one of the next steps.

The ACD-B and DE-F are simplifications of Dünnwalds 'nasality' and 'clarity' parameters extracted from wider parts of the spectra. The L parameter we explained in the start of this thread. The SPL/Force is simply the overall sound level of the entire spectrum.

[Anders Buen]

Here is what Dünnwald say about the cumulative distribution plots of the L parameter:

Alle vier Kurven haben einen ahnlichen Verlauf, jedoch der Mittelwert ist von den Fabrikviolinen a ) uber die Meisterinstrumente b ) und die alten Meistergeigen

c ) zu den altitalienischen Violinen d ) stark zu hoheren Werten von L verschoben. Alle sehr guten Instrumente des Qualitatsbezugs liegen bei Werten

von mehr als 18 dB. Die Aussagekraft dieses Parameters ist erstaunlich gross, denn nur 34% aller Instrumente konnen mit dem Qualitatsstandard konkurrieren,

66% sind nicht mehr akzeptabel. Das bedeutet jedoch nicht, dass alle Instrumente mit einem Wert L > 18 dB sehr gut sind. Es gibt einen Teil unter diesen, der zu den als besonders schlecht bekannten Instrumenten gehort und ebensoviele nur mittelgute Violinen. Somit sind weitere Grossen notwendig, um sehr gut klingende Violinen eindeutig zu charakterisieren.

He basically say that all very good violins has L parameter values above 18 dB, but there are instruments with L >18 dB that are known to be especially bad. There is also a group of instruments with L > 18 dB that are only characterized as being mediocre or 'medium good'.

Therefore it is necessary to extend the number of factors to better characterize the very good sounding instruments.

In that process he divide the spectrum into frequency regions from A to F as seen in the enclosed figures. He found that the sum of the sound pressure levels in the A, C and D was beneficial, and that the B region was having a negative influence on the timbre if it is strong. So the parameter became ACD-B and he denote that instruments with B stronger than the sum of ACD regions would sound more Topfig or nasal as given in his CASJ paper.

He also found that strong response in the the high frequency region D and E was beneficial while the very high frequency region F (harshness region) should be weak. The CDE region is the 'body hill' - 'singers formant' brilliance and carrying power region. He made a parameter combining the levels of the D and E regions and subtracted the level in the harshness frequency range F: DE-F He connect higher values of this parameter to correlate with a clearer or warmer (?) sound (Der Ubergang von den Bereichen D und E zu F fuhrt zu klaren oder heiseren Klangen, je nach Grosse des Abfalls des Spektrums von DE nach E) (Heiser translates to: gravelly, hoarse, croak, hoarsely, rauchig, husky, huskily I do not know the correct translation to use here nor what exactly Dünnwlad mean by his phrase except for a weak F compared to the D and E regions seem to be better). With a strong F in relation to the DE region the sound may be said to be less clear and husky wich is not good.

He did the summation of sound pressure levels for each played note and calculated a percentage of un-nasal (higher ACD-B values) and clear (higher values of DE-F values) sounds. The old Italians had higher percentages of such notes than the rest and he improved the prediction of the very good sounding violins from their spectra from what he got from just the L parameter alone. He also got some valuable information as to which notes would sound better than others. There may be such less good notes even in the best instruments, but there may be fewer of them than in the more mediocre instruments.

[StuartRochon]

Is there a good book that explains all this?? I have a hard time figuring out which peak is the A0 and B1 just cause I don't know enough. Are there any reference books of the bridge taps for the great instruments?? Where is it good to have peaks and where is it bad to have peaks?

[Anders Buen]

Is there a good book that explains all this?? I have a hard time figuring out which peak is the A0 and B1 just cause I don't know enough. Are there any reference books of the bridge taps for the great instruments?? Where is it good to have peaks and where is it bad to have peaks?

Unfortunately there is no good book that explain all this except for Janssons book on violin acoustics http://www.speech.kth.se/music/acviguit4/index.html, the Strad 3D project DVD, Joseph Curtins Strad article on modal data from Strads and del Gesus, I do not remember which issue.

Don Noon wrote a fine article on the subject which he published here in one thread a while ago.

[violins88]

Unfortunately there is no good book that explain all this except for Janssons book on violin acoustics http://www.speech.kth.se/music/acviguit4/index.html, the Strad 3D project DVD, Joseph Curtins Strad article on modal data from Strads and del Gesus, I do not remember which issue.

Don Noon wrote a fine article on the subject which he published here in one thread a while ago.

Both articles by Don can be see on my website.

here

Click on "violin mode measurements 1" and (2).

[Don Noon]

Anders,

Have you found any parameters that correlate to a high DE-F value?

One source of potential gold in your data mine is exploring parameter combinations. Specifically, the B1+ frequency would appear to be to be influenced by a combination of back center bout thickness, back crossgrain modulus, bassbar stiffness, and top longitudinal stiffness. The correlation of B1+ with one parameter might not show very strong correlation due to the scatter in the other parameters, but it might be possible to derive a combination that would have much better correlation. I imagine there are algorithms that might do this statistically (your specialty?), but it is also possible to make an educated guess from knowing the mode shape and simple structural analysis (my specialty). Are you interested in this, or have you done it already?

Do you have damping in your database?

[Wm. Johnston]

When it comes to Curtins idea of along grain to across grain stiffness, that can e.g. be measured with compression sound speeds along and across the plates.

Seems like this would be strongly affected by the arching of the plate.

[Anders Buen]

Anders,

Have you found any parameters that correlate to a high DE-F value?

Yes, f-hole length and arch height come out in multiple regression as being significant. But also the transversal sound speed of the back plate, and the sound speed ratio of the top plate correlate. The higher the cross grain sound speed of the back, the weaker the DE-F. This may be related to properties of the hardangers. So the effects here are complicated as seen in the scatterplots. E.g. the effect of f-hole lengh is different for the hardangers and the violins. You also see that in general hardangers have higher values.

Here we see an example of different correlations for the hardangers and the violins regarding the f-hole length. A data separation between hardangers and violins might be necessary at some point here.

I think that in general the body hill frequency tend to be lower in the hardangers, possibly due to the longer f-hole wings and thus lower flapping frequencies. A lower tuned in plane bridge mode might also be an effect here. The central region of the top plate also tends to be thicker in hardangers to compensate for the flatter cross grain arch there.

The color plot here is somewhat risky to follow for a violin maker. It seem to fit better for us working on hardangers, though. The data are preliminary.

One source of potential gold in your data mine is exploring parameter combinations. Specifically, the B1+ frequency would appear to be to be influenced by a combination of back center bout thickness, back crossgrain modulus, bassbar stiffness, and top longitudinal stiffness. The correlation of B1+ with one parameter might not show very strong correlation due to the scatter in the other parameters, but it might be possible to derive a combination that would have much better correlation. I imagine there are algorithms that might do this statistically (your specialty?), but it is also possible to make an educated guess from knowing the mode shape and simple structural analysis (my specialty). Are you interested in this, or have you done it already?

Yes I started off looking for a good multiple regression model for the B1+ frequency, the results was given in another tread here a couple of weeks ago. http://www.maestronet.com/forum/index.php?...st&p=457079 The best I could find that time was based on central back plate thickness and mode 5 top plate frequency. In general it is easier to find correlations for the mode frequencies than the levels. I have more data on the frequencies, from some 55 instruments now than for the levels (31 instrument entries).

Other factors influencing the B1+ frequency: Border thickness of the top plate, top plate weight, mode 2 and 5 frequencies, bassbar lengh, bassbar upper height, mean border thickness of the back plate, mean thickness of the back plate, maximum thickness of the back plate (all these correlate strongly between them), and the "Harris stiffness" of the top plate.

Do you have damping in your database?

Not included in the data mining spreadsheet. But it is possible to extract damping for the modes from the measured wav files of the impacts. I think it is easier to read out the damping from the admittance (accelerometer) measurements than the sound pressure signal, as the admittance files are cleaner. But the damping of higher modes is more difficult to read out. Have you tried the "reverberation method" using e.g. one third octave filters? That probaly has to be done on data from an accelerometer, as the room acoustics will affect the reverberation measured by a mic, at least if the room reverberation is longer than the wooden rev. With some luck maybe the room reverberation algorithm in SpectraPlus can do that on the accelerometer signal from the decaying impacts. Would expect shorter rev times in the violin wood than normal for room reverberation.

[Anders Buen]

One source of potential gold in your data mine is exploring parameter combinations. Specifically, the B1+ frequency would appear to be to be influenced by a combination of back center bout thickness, back crossgrain modulus, bassbar stiffness, and top longitudinal stiffness. The correlation of B1+ with one parameter might not show very strong correlation due to the scatter in the other parameters, but it might be possible to derive a combination that would have much better correlation. I imagine there are algorithms that might do this statistically (your specialty?), but it is also possible to make an educated guess from knowing the mode shape and simple structural analysis (my specialty). Are you interested in this, or have you done it already?

Any input that can improve prediction or the understanding of the subject is interesting, Don. With lots of measured data one may state 'questions' to them, and the smarter the 'questions' the better the 'answers' may be. I am interested in statistics, but have not used it 'professionally' since more than 10 years ago when I was mining on predicting, and possibly how to improve, the current efficiency of aluminium electrolysis cells based on monitored process parameters. However, I have been fiddling with statistical methodes for a while on subjects like the one we see here, initially combining spectral data extracted from the Miracle Makers recording with Oliveira and instrument data from Biddulph et als book on del Gesus + a few other sources that are played on that recording.

I do not really know what my 'speciality' is. I have an interest for reading literature on the violin acoustics subject, I enjoy doing measurements, and looking at and presenting data, I am interested in the sound of the instruments and enjoy making and repairing them when the sufficient energy and focus is there. I also enjoy solving or working on complex problems. I aspire to do finite element analysis on violins and other structures. John Masters initially got me up and going on that, using his models, but I haven't focussed on that for a while. I wish I could as a part of my profession. A few year ago I made a simple model for predicting signature mode frequencies and levels in violins based on Morals PhD empirical statistics model work and modified flat plate theory. I haven't finished off that model and article yet. All the other work I do, like this 'data mining thing', might give different approches to the same problem that hopefully may 'line up' into similar conclucions and a consistent 'puzzle picture'. What I present here is one part of that approach.

Maybe we are looking for patterns that are not really there? But I would believe that many makers do beleive they see one, or several of them. I think I do see something of a 'pattern' and hopefully it will grow clearer and more detailed or maybe there might be some surprises waiting? Recognition of patterns and 'modeling on them' is one of our brains specialities, along with its creativity and ability to cope with complex information. The best 'data miner' is probaly our brain. But it can live and work with models that are fuzzy, partly or completely unconscious and need only to be sufficient for its purpose to keep us going at what we do at a certain degree of 'economy', comfort and effort. A model that may work well for some or just even just one person, may not be sufficient for another. But still it may do its job for the single person. I think we see a lot of 'fuzzy models' that seem to work more or less efficiently in the violin making.

Many brains may work more efficiently than just one on solving a complex problem. That is one of the reasons why I am here posting results or data and discussing instead of putting it into articles that may delay years before they appear available to others. A lot of interesting information will never be published either due to limited success, lack of clear patterns or conclusions, etc. My data mining work on the Miracle Makers recording is one example of that. The CASJ nor the VSA papers rewievers would not recommend to publish that. So therefore I have many data mining sets I work on. Thay may give a better complete picture than just one would. What I present here is data from one out of five such sets I work on. In principle it is possible to make one for each recording like the MM, or the Glory of Cremona, or the Ehnes recording on Fultons violins, or Riccis old 'The Violins of Cremona' when instrument data from these recordings become available. There are some obviolus benefits from working on instruments that one can lay the hands on and measure in any form at will.

I am grateful for any input or ideas that can lead to improvements in our understadning. So, Don if you have some more ides on what to look for, I am ready to include that into my work.

[Anders Buen]

Here is a contour plot including the data points for the B1+ frequency versus top plate mode 5 tap tone and back plate average central thickness.

I have excluded the most extreme values in both ends, but for the comparison I also include the plot including these. The trend remain the same, but the color scale and intervals are different.

[Anders Buen]

Spectral balance between the signature modes is one of Janssons earlier findings for instrument quality. I have had problems with the B1+ modes being weak in some of my hardangers, so the reason why has interested me to find out. Here is one factor that may play a role, the f-hole length.

In the Pietro Guarneri example of Dünnwalds figures in my first post in this topic you see that the levels of the B1- and B1+ modes are about balanced. For the Strad example the B1- is a bit stronger than the B1+, but not by large. Often the trend is opposite in fine old violins.

From this figure we may assume that the many Ole Bull copies out there might turn out to have a stronger B1- than the B1+ because the f holes are long in that model. Is that the case? And how about the original?

[Don Noon]

Whenever there appears to be an influence of a construction feature on the sound (in this case, long F-holes giving stronger B1- levels), I want to know why. So I spent a while re-pondering the Strad3D animations, and arrived at a plausible hypothesis. Or two.

The first one is simple: longer F-holes might raise the A0 frequency to where it has some amplifying effect on B1-. I kinda doubt this, and it's easy to check: are higher A0 frequencies correlated to the longer F-holes? Are higher A0 frequencies correlated to stronger B1- regardless of F-hole length?

The second one is more complex: the B1- is roughly characterized by longitudinal bending of the back plate and crossgrain bending of the top. Longer F-holes will weaken the crossgrain stiffness of the top, thereby making the back stiffness more influential in the total mode stiffness. Energy losses are much higher in crossgrain bending of spruce than in longitudinal bending of maple, so longer F-holes should make the B1- mode less lossy, therefore higher amplitude.

This hypothesis can be crosschecked a number of ways: longer F-holes should correlate with lower B1- frequency. Lower crossgrain speed of sound in the top should have a similar effect to long F-holes, i.e. stronger B1-. A thinner top as well should strengthen B1-. I'm not claiming any of these as great predictions... this is just a list of things you should find IF my wild hypothesis is to hold any water.

[Anders Buen]

Whenever there appears to be an influence of a construction feature on the sound (in this case, long F-holes giving stronger B1- levels), I want to know why. So I spent a while re-pondering the Strad3D animations, and arrived at a plausible hypothesis. Or two.

The first one is simple: longer F-holes might raise the A0 frequency to where it has some amplifying effect on B1-. I kinda doubt this, and it's easy to check: are higher A0 frequencies correlated to the longer F-holes?

Thanks for doing this Don. The A0 frequency does not appear to correlate with long f-holes in this set. The intuition would be that it should have, but maybe the longer f-holes tend to have a narrower opening.

[Anders Buen]

Are higher A0 frequencies correlated to stronger B1- regardless of F-hole length?

Sort of, but not a significant correlation though statistically. But we would suppose a correlation here by the underlying physics. So that would account for accepting that there is a trend to be expected here.

[Anders Buen]

The second one is more complex: the B1- is roughly characterized by longitudinal bending of the back plate and crossgrain bending of the top. Longer F-holes will weaken the crossgrain stiffness of the top, thereby making the back stiffness more influential in the total mode stiffness. Energy losses are much higher in crossgrain bending of spruce than in longitudinal bending of maple, so longer F-holes should make the B1- mode less lossy, therefore higher amplitude.

That intuition seem to have support in the data, especially for the violins. There is an outlier in the hardanger data here that seem to screw up the correlation there. Interesting observation.

[Anders Buen]

Lower crossgrain speed of sound in the top should have a similar effect to long F-holes, i.e. stronger B1-.

No clear correlation here, but an interesting possible category difference between a group of high B1- level violins versus a low group. Why?

There is sort of paralell lines increasing trend lines in the violin data, in correlation with your intuition here.

Need to go to work now. I'll pick up on this tonight. Thanks this far!

[Don Noon]

No clear correlation here...

Hidden in the original assumption is that damping is independent of the speed of sound. That got me thinking... it seemed to me that lower density wood (which is strongly related to higher speed of sound) measured higer damping. So I gathered my data and plotted it. This is all from a single Sitka board and a single bigleaf maple board, except for the two outliers that were from wood a friend of mine gathered in the forrest somewhere.

If you look at all the data together, there is an obvious correlation between high speed of sound and low damping (i.e. high Q factor).

Zooming in on only the crossgrain spruce, the correlation isn't so obvious, but I think there is one (I don't know how to do those fancy statistical trendlines; I just look at the dots for a general slope). Even a slight slope could reverse the idea that lower crossgrain C in the top would give lower overall damping in B1-. In any case, the scatter is huge... and this is just for a single board from one type of spruce.