Head-Fi.org › Forums › Equipment Forums › Sound Science › What makes piano sound so hard to reproduce?
New Posts  All Forums:Forum Nav:

What makes piano sound so hard to reproduce? - Page 11

post #151 of 191
I'm patient, but xnor can't reveal the answers to his test until he gets a chance to take it. There are other people waiting for the answer to their tests too.
post #152 of 191
Quote:
Originally Posted by jaddie View Post

There are adjustments for all physical alignments on the head and stylus, and yes, azimuth is there.  There are optical methods used to adjust the stylus position, but what you have to realize is this is one of those things that you can obsess about all day, but when it comes down to it, the precision to which the cutter stylus is aligned far exceeds the precision to which a turntable is usually aligned. The fact that you've taken the trouble to play test records that show up az issues (you're probably looking at channel separation, right?) is very unusual, and yes, that would show up misalignments like that.  But it's actually sort of a hyper sensitive test.  In practice, you can't actually hear small changes in channel separation once they're up where they belong.  What's odd about your observation is that test records weren't exactly made in the mass-quantity of a popular release. I would think there would be very little need to cut a fresh lacquer. You can get about 1000 pressings out of a stamper, and 8 to 10 stampers out of a mother, and about the same number of mothers out of a father, so all together they should have gotten something like 10,000 records before a new master would be needed, and if they did the 3 step process, something like 100,000 if memory servers.  That's a lot of boring test records! But possible they underestimated, and did the two-step process, got their 10,000 copies, then the record went platinum and the had to cut new masters.

 

Anyway... though I haven't surveyed labs to that detail, I would expect that once the stylus is aligned to a certain precision, they stop worrying about it.  Every change in the head and stylus would dictate a realignment, which the responsible labs would have done, to the best of their ability. But their reference is an optical process used to determine a physical position.  They couldn't use a playback cartridge to confirm it because that cartridge and stylus has the same alignment problem, so there are limits to how precise this adjustment can be.

 

As to industry or my personal practice in lathe alignment, full disclosure, I was not responsible for any lathe maintenance.  I was a client of the lab, not an employee.  Yes, I'm an engineer, but for those projects was hired to a rather odd position, that of technical supervision.  I was one of the engineers who designed and built the studio where the projects were recorded, and assisted during the sessions, then the record company hired me to track and supervise the projects technically all the way to production of the final releases, which were both vinyl and CD.  My solution to precision lathe calibration, and in fact, the entire analog mastering chain, was to work with the best lab I could find, interview them thoroughly, and then trust their team to make sure the gear was in peak condition.  Then I supervised the actual mastering sessions, observed and confirmed the operation of everything (contributed very little!) and left with the reference lacquers under my arm. I didn't supervise the process at the making of the metal parts or the process pressing plant, but again, hired the best I could find, and then beat them all up over bad test pressings anyway.  The process past cutting the lacquer is really difficult, and full of problems.  We went through several different vinyls, one bad mother/stamper combo, etc. At least half a dozen test pressings before we got a good one.  But right up to cutting the lacquers, it's pretty well in control.  Side note,  by contrast, the CD version came back perfect the first time.

 

And now, to put this "on topic"... our piano sound on those records was superb.  The Steinway D never sounded better. Whew.

Wow - so you seem to be the right guy to ask such things indeed.

 

I could not agree more with you that test records are boring - but in a sense, they are the same as saying : There is no more practical thing as a good theory ! I certainly do not enjoy aligning each and every cart until it is "perfect" ( tedious, time consuming, nervous ( like n-times repeating a move that can wreck an expensive stylus if just slightly wrong ... ) - in a way, bore to the max ) - but once done, the sonic results are worth it !

 

I did not mean with differences in azimuth on the same type of test disc - but same brand products at least 10 years apart between the first and second type of test disc. That would certainly have to use another cutting stylus - and be subject to slight azimuth misalignment, just as you have described. No pick up arm/cartridge can be so precisely made to improve on optical alignment - but that is not absolute in accuracy, it finally does boil down to operator's sight and decision when it is "good enough".

 

In the late 50s, an South American engineer published in JAES a theorethical paper regarding azimuth and ultimate channel separation of which a stereo record is capable of. Just shy of 60 dB, something between 55 and 60 IIRC. The error in angle that would be detrimental is in mninutes, not degrees ! I thought that this can never be achieved in real life. But, Allearts does publish 60 or even more dB channel separation for his better carts.

Still, I thought this must be fairy tales. Up to that point, my best figures were symetrical separation both l>r and r>L just above 40 dB across the entire  20-20 K range, or slightly less above 10 kHz, but still above 30-35 dB at 20 KHz. With one VERY selected Benz TR cartridge on Versa Dynamics 2.3

record playing system it was finally achieved - 60 or so dB separation, it was at or below the noise of the JVC TRS 1007 test record. I will check if I still keep the DSD recording of that setup session. Not mine table/cartridge, but friend's - after hearing that sonics, nothing else would do for quite some time after hearing it. I am certain I have no musical recordings - phono preamp used had so great gain  ( at +18 dB @ 300 Hz 10 V RMS was coming out of that thing !! ) that overloaded my DSD recorder - the majority of measurements was at 0 dB and they were just within the overload of the recorder input, music that can go as high as + 18 dB of course overladed the input. So - no music, recording of measurements only ( if they survived ).

 

Above proves there are azimuth and precision in general fanatics out there ...

 

Considering the patience and devotion with which you obviously went about lacquer mastering, I bet "your" piano DOES sound good !

 

Any more info on record cutting most welcome !

post #153 of 191
Quote:
Originally Posted by xnor View Post

analogsurviver, can you hear the difference (<- that's a link to a small listening test)?

Answer in the "can you hear the difference" thread.

post #154 of 191
Quote:
Originally Posted by nick_charles View Post

 

You could also just use FooBar with the ABX plug-in - much simpler and also truly double-blind 

I like foobar - but DACs in Korg MR series so much outstrip my PC that it does not make sense to use whatever PC I have. Perhaps when I add Mytek

192/DSD DAC - but that is 1K6 Euro and with no way to hear it prior to purchase, I "think" it will have to wait a bit. I would not like to discover all the difference would be the convinience to play DSD directly using PC and HD, without having to upload files to Korg MRs - the SQ should be better as well for that kind of money.

 

I generally do not rely on ABX of short samples - in the long run, I prefer listening for longer periods of time, as finer nuances usually do not come across in short ABX tests - which are nonethles useful . I like to listen like that on speakers while doing chores around the house - and if it does not disturb or bother me in some way, - it must be really good.

post #155 of 191
Quote:
Originally Posted by analogsurviver View Post

Wow - so you seem to be the right guy to ask such things indeed.

 

I could not agree more with you that test records are boring - but in a sense, they are the same as saying : There is no more practical thing as a good theory ! I certainly do not enjoy aligning each and every cart until it is "perfect" ( tedious, time consuming, nervous ( like n-times repeating a move that can wreck an expensive stylus if just slightly wrong ... ) - in a way, bore to the max ) - but once done, the sonic results are worth it !

 

I did not mean with differences in azimuth on the same type of test disc - but same brand products at least 10 years apart between the first and second type of test disc. That would certainly have to use another cutting stylus - and be subject to slight azimuth misalignment, just as you have described. No pick up arm/cartridge can be so precisely made to improve on optical alignment - but that is not absolute in accuracy, it finally does boil down to operator's sight and decision when it is "good enough".

 

In the late 50s, an South American engineer published in JAES a theorethical paper regarding azimuth and ultimate channel separation of which a stereo record is capable of. Just shy of 60 dB, something between 55 and 60 IIRC. The error in angle that would be detrimental is in mninutes, not degrees ! I thought that this can never be achieved in real life. But, Allearts does publish 60 or even more dB channel separation for his better carts.

Still, I thought this must be fairy tales. Up to that point, my best figures were symetrical separation both l>r and r>L just above 40 dB across the entire  20-20 K range, or slightly less above 10 kHz, but still above 30-35 dB at 20 KHz. With one VERY selected Benz TR cartridge on Versa Dynamics 2.3

record playing system it was finally achieved - 60 or so dB separation, it was at or below the noise of the JVC TRS 1007 test record. I will check if I still keep the DSD recording of that setup session. Not mine table/cartridge, but friend's - after hearing that sonics, nothing else would do for quite some time after hearing it. I am certain I have no musical recordings - phono preamp used had so great gain  ( at +18 dB @ 300 Hz 10 V RMS was coming out of that thing !! ) that overloaded my DSD recorder - the majority of measurements was at 0 dB and they were just within the overload of the recorder input, music that can go as high as + 18 dB of course overladed the input. So - no music, recording of measurements only ( if they survived ).

 

Above proves there are azimuth and precision in general fanatics out there ...

 

Considering the patience and devotion with which you obviously went about lacquer mastering, I bet "your" piano DOES sound good !

 

Any more info on record cutting most welcome !

Well, 60dB seems a lot to hope for.  The practical view, though, is that you might do that on a specific test record, but it's highly unlikely that kind of crosstalk (or lack of it) would be present on all or even most records.  You can always achieve incredible performance when you align the play system to the record system, but the problem is, it's very unlikely that two record systems will match. In fact, speaking of test records, try measuring response using two different test records where the sweep is recorded with RIAA eq.  You'll find they don't agree much.  Which is why when I set up turntables I use a test record where the response sweep is recorded without RIAA eq (constant velocity), then match the result to an ideal curve.  In any case, you see the issue.  You set up to get that kind of crosstalk, then play a different record where the cutter wasn't quite set the same, and you're back to 35 - 40dB which is the practical and stable limit.  

 

Speaking of crosstalk,  were you aware that cutting systems use dynamic control over the vertical groove depth? It's a means of controlling vertical groove distortion and over-cutting since the vertical (depth) has less total modulation space (too deep you hit the aluminum, to high you've got no groove left), but the result is a dynamically variable amount of crosstalk!  Taken to extremes, if you totally kill the vertical component, you're cutting in mono.  So as active dynamic vertical control works, crosstalk is varied anywhere from the theoretical max to somewhat less than the typical 35dB.  Ok, a lot less on some peaks.  Vertical groove limits are one of the reasons kick drums are mixed panned center. 

 

So tweak the cart for 60dB on a test record, and in cutting we'll knock you down to 20dB or less on a moment by moment basis.  

 

I hope that one doesn't make your head explode, but I'm going to duck and cover anyway.

 

I wish the piano was mine...it wasn't, it belonged to the studio, actually I think it was a lease.  Tuned weekly, though. Very sweet to play.


Edited by jaddie - 1/3/13 at 3:29pm
post #156 of 191
Quote:
Originally Posted by jaddie View Post

Well, 60dB seems a lot to hope for.  The practical view, though, is that you might do that on a specific test record, but it's highly unlikely that kind of crosstalk (or lack of it) would be present on all or even most records.  You can always achieve incredible performance when you align the play system to the record system, but the problem is, it's very unlikely that two record systems will match. In fact, speaking of test records, try measuring response using two different test records where the sweep is recorded with RIAA eq.  You'll find they don't agree much.  Which is why when I set up turntables I use a test record where the response sweep is recorded without RIAA eq (constant velocity), then match the result to an ideal curve.  In any case, you see the issue.  You set up to get that kind of crosstalk, then play a different record where the cutter wasn't quite set the same, and you're back to 35 - 40dB which is the practical and stable limit.  

 

Speaking of crosstalk,  were you aware that cutting systems use dynamic control over the vertical groove depth? It's a means of controlling vertical groove distortion and over-cutting since the vertical (depth) has less total modulation space (too deep you hit the aluminum, to high you've got no groove left), but the result is a dynamically variable amount of crosstalk!  Taken to extremes, if you totally kill the vertical component, you're cutting in mono.  So as active dynamic vertical control works, crosstalk is varied anywhere from the theoretical max to somewhat less than the typical 35dB.  Ok, a lot less on some peaks.  Vertical groove limits are one of the reasons kick drums are mixed panned center. 

 

So tweak the cart for 60dB on a test record, and in cutting we'll knock you down to 20dB or less on a moment by moment basis.  

 

I hope that one doesn't make your head explode, but I'm going to duck and cover anyway.

 

I wish the piano was mine...it wasn't, it belonged to the studio, actually I think it was a lease.  Tuned weekly, though. Very sweet to play.

OK,  I am perfectly aware there is a LOT mumbo jumbo going on when mastering a lacquer disc - it is always a compromise of playing time vs quality

and if you look at the requirements for the master tape/file ment to be for LP release, In general, requirements limit the amount of bass and amount of vertical modulation - that there is some manipulation to the vertical groove depth I was aware of, but dynamic one is new to me - although it perfectly makes sense. I meant to wrote another "wall" of text regarding reproducing the LP ; it has a similar set of limitations as mastering, but I propose we continue "washing dirty analog loundry" in PMs - who knows, we might come up with a solution that might yield an audible improvement for the general public who just puts a LP on the turntable and wants to listen to some music. I always try to act as the one bringing it all together - if one insists "this is a problem of reproduction" and another "this is a problem of mastering", it will never get better - working together usually yields an overall improvement.

 

These facts are reflected in channel separation of the better  phono cartridge - they hover around 35 dB, which is also the value measured with Scanning Electron Microscope ( SEM ) of the JVC TRS 1007 test record, basically confirming your claims. That holds true for static signals ( sweeps ) 

used, with real music and dynamic vertical groove depth control ...

 

Said limitations can never be ameliorated to the point of being equal with GOOD digital. I found the only digital currently available to general public to unite most of the positive attributes of analog ( be it tape or vinyl record ) and "digital" is DSD 1 bit at 5.6 MHz sampling frequency. SACD is DSD 1 bit at 2,8 MHz - although it is better than CD, it is not THAT superiour to CD not to have enough of CD diehards to stop it - which basically happened. At 5.6 MHz, this game is over before it begins; trouble is, there is no commercially available medium or DSD downloads at 5.6 MHz - only 2.8 MHz. In view of this, super carefully half speed mastered analog record from either analog master tape or DSD at 5,6 MHz with frequency response about equal with that of DSD when played by top phono cartridges is still viable - despite the fact that I am perfectly aware it is 2013 now.

 

I unfortunately can not play any instrument, but "your" good piano tuned weekly must have been balm for the ears. A regularly tuned and well taken care of  piano of relatively less known/appreciated manufacturer can in fact be superiour to the big name one that is less well taken care of.

post #157 of 191
Quote:
Originally Posted by xnor View Post
I read somewhere that the high-frequency noise can cause distortions in the audible frequency range, but this sounded more like a hypothesis than anything else.

Because of intermodulation distortion, this actually does occur to some extent.

post #158 of 191
Quote:
Originally Posted by spaark View Post

Because of intermodulation distortion, this actually does occur to some extent.

Unfortunately true. This is perhaps one of the reasons why phono cartridges with almost indistinguishable performance in the audio band can sound markedly different. Cartridges with high Q (undamped) resonance anywhere in their response, be it so ludicrously outside audio band, are more likely to cause trouble than the ones that show reasonably smooth response throughout their entire operating range of frequencies. In 1980 or so, Peter Moncrieff of International Audio Review published a comparative survey of then pretty much any cartridge then jostling for the place on audiophile's tonearm : frequency limit for the measurements were set by the equipment ( spectrum analyzer ), which at the time could not measure beyond 256 kHz ( in a word: two hundred  fifty six kilohertz ). No test record can go so high, a pulse response ( simply lowering the stylus to the glass support, creating vertical pulse ) was measured. The first in the series of Dynavector Karat MC carts, the 100R Ruby and 100D Diamond ( respective materials cantilevers of these carts were made of ) plus some others could easily exceed the measurement limit. 100D was the best behaved of the lot in this test. Today's Dynavector DV17D3 , the latest version of the venerable 100D which has by present standards become middle priced cartridge, should be a bit faster / better still. Mr.Moncrieff performed a lot of unconventional measurements, which he felt represent the actual behaviour of cartridges in real life better than conventional measurements of requency response and channel separation and provide reasonably good correlation between objective measurements and subjective evaluation by listening. Those seriously interested in analog vinyl might try inquiring if this survey is still available : http://www.iar-80.com/

post #159 of 191
Quote:
Originally Posted by spaark View Post

Because of intermodulation distortion, this actually does occur to some extent.

Let's take a look at what intermodulation distortion is and see if this can be the mechanism.  IMD happens when one frequency modulates another causing a third result.  There are three basic types.  One, where low and high frequencies interact.  An example would be bass notes modulating mid-band or high frequencies because of non-linearity in the system.  In testing for this, an IMD analyzer typically uses the SMPTE standard of 60Hz and 7KHz mixed 4:1, and looks for whatever other frequencies are produced.  The audible nature of this kind of distortion is audible as the effect of low frequencies modulating higher ones, which is hard to describe, but easy to hear.  Find the loudest FM station on the dial and listen. Their on-air processing produces quite a bit of this kind of distortion.

 

The second type is difference IMD, where two closely spaced frequencies mix non-linearly to produce new frequency as products of the two.  For example, two tones of equal level, one at 10KHz, the other at 11KHz, may produce a 1KHz tone, which would be the low frequency second order product.  The non-linearity that produces this kind of result would also produce a high frequency product, and possibly other order products.  If one tone is of a significantly different level than the other, the IMD products change in level according to a non-linear relationship.  For example, if the two tone were 10dB apart, the products would likely be reduced by much more than 10dB.

 

The third type of IMD is transient-induced (TIM) or slew-induced intermodulation distortion (SID) or dynamic intermodulation distortion (DIM).  Three names for the same thing.  When a device is pushed to the speed limit of its ability to change output voltage levels, it's said to be slew-limited.  While it's slewing to a different output voltage, it looses its ability to output anything else but the changing voltage.  This means that during the transition all other audio will be essentially modulated downward until the transition is complete.  The typical test was to mix a mid-band square wave with a high frequency sine wave.  The exact frequencies depend on the bandwidth of the system being measured.  The output spectrum is then compared with the input (undistorted) signal, and the level of new frequency products is noted.  The technique was presented in the paper "A Method for Measuring Transient Intermodulation Distortion (TIM)" by Leinonen, Otala and Curl, J-AES April 1977.  Subsequently, it was found that amplifiers with proper negative feedback design reduced or eliminated the TIM effect, and that coupled with the poor correlation between measurement and audibility caused this test to fall into disuse. 

 

The points to note here are: for IMD of any kind to take place, the system has to have some degree of non-linearity, and at least one frequency component has to push the system well into that non-linear region. 

 

So, what IMD mechanism would cause high frequency noise to cause distortion products to land in the audible range?  We can eliminate the first type, LF modulation HF, because what's suggested is in fact the reverse.  The spectrum and amplitude of the noise is unspecified, but it's a fair assumption that the noise is fairly low level and above the audio band.  We can thus eliminate the second type, as the noise is non-specific, low level, and above the audio band.  If the noise forced a system into non-linear operation to produce intermod products, they would relate in frequency to the audio and the noise, placing the products out of the audio band as well.  To put it another way, for IMD to be audible the resulting products must be in the audio band, and the result of IMD of low level high frequency noise and audio frequencies would be either out of the audio band or extremely low level.  

 

It's not looking to me like high frequency noise would cause audible problems because if intermodulation.

post #160 of 191
Quote:
Originally Posted by jaddie View Post

So, what IMD mechanism would cause high frequency noise to cause distortion products to land in the audible range?  We can eliminate the first type, LF modulation HF, because what's suggested is in fact the reverse.  The spectrum and amplitude of the noise is unspecified, but it's a fair assumption that the noise is fairly low level and above the audio band.  We can thus eliminate the second type, as the noise is non-specific, low level, and above the audio band.  If the noise forced a system into non-linear operation to produce intermod products, they would relate in frequency to the audio and the noise, placing the products out of the audio band as well.  To put it another way, for IMD to be audible the resulting products must be in the audio band, and the result of IMD of low level high frequency noise and audio frequencies would be either out of the audio band or extremely low level.  

 

It's not looking to me like high frequency noise would cause audible problems because if intermodulation.

 

Interesting, when researchers tried to follow-up the controversial Oohashi experiment which used a Balinese Gamelan (capable of very high harmonics at above -70db all the way up to 40khz) they were unable to replicate the findings. They suggested that  the interaction of the rather specialized super-tweeter and other drivers created IMD that influenced listeners cortices but was not "hearable" are you saying that explanation does not hold water even with something s extreme as the gamelan ? Most researchers consider the Oohashi work as largely discredited due to this explanation.

post #161 of 191
Quote:
Originally Posted by nick_charles View Post

 

Interesting, when researchers tried to follow-up the controversial Oohashi experiment which used a Balinese Gamelan (capable of very high harmonics at above -70db all the way up to 40khz) they were unable to replicate the findings. They suggested that  the interaction of the rather specialized super-tweeter and other drivers created IMD that influenced listeners cortices but was not "hearable" are you saying that explanation does not hold water even with something s extreme as the gamelan ? Most researchers consider the Oohashi work as largely discredited due to this explanation.

I've always had trouble with the "not hearable" influencing perception. 

 

(say Hi to Nora and Asta for me)

post #162 of 191

@jaddie: I didn't quite understand your reasoning as to why the second type of IMD doesn't apply.

post #163 of 191
Quote:
Originally Posted by spaark View Post

@jaddie: I didn't quite understand your reasoning as to why the second type of IMD doesn't apply.

The second order product would be at a frequency that corresponds to the difference of the two frequencies involved.  If we're talking about high frequency noise in high bandwidth phono systems, where would that noise be?  Probably rising in level at 6dB/octave, so more as you go up.  Let's say the spectrum around 50KHz is "noisy", and we're looking at a fairly high frequency in the audio band...what the heck, let's go for 20KHz.  The difference is 30KHz, still out of the audible spectrum.  And that assumes both frequencies are pushing well into serious non-linearity, which even though noise would rise at 6dB/octave, it's still...welll....noise.  That puts it at a low level, though in this discussion we've never assigned a specific frequency or level.  

 

Before we start going for a lower noise frequency to put the intermod products in band, remember, the noise level drops as you move down in frequency. Lower the level of one intermod source frequency and the products lower too, only faster because its a non-linear system.

 

Also, if broad spectrum noise is one component, then the intermod product will also be broad spectrum noise.

 

I guess before we continue the discussion somebody should specify something about the noise.

post #164 of 191

On the subject of reproducing piano sound, this is something I care about not least because as a lover of Debussy and Ravel (amongst others) I listen to a lot of piano!

 

A great advantage headphones have over regular speakers is that they have no cross-over. I think that the cross-over is often a very problematical area in loudspeakers. The piano frequency range often straddles the cross-over and when you are listening to a particularly intense piece of piano playing you become focussed on this area of the frequency range.

 

I think a good headphone setup can produce very good piano playing however a good amplifier really helps. To capture all the detail, those varying rhythms, the subtleties of tones etc. I suggest a good class A solid state amplifier allied to transparent headphones.

post #165 of 191
Quote:
Originally Posted by jaddie View Post

The second order product would be at a frequency that corresponds to the difference of the two frequencies involved.  If we're talking about high frequency noise in high bandwidth phono systems, where would that noise be?  Probably rising in level at 6dB/octave, so more as you go up.  Let's say the spectrum around 50KHz is "noisy", and we're looking at a fairly high frequency in the audio band...what the heck, let's go for 20KHz.  The difference is 30KHz, still out of the audible spectrum.  And that assumes both frequencies are pushing well into serious non-linearity, which even though noise would rise at 6dB/octave, it's still...welll....noise.  That puts it at a low level, though in this discussion we've never assigned a specific frequency or level.  

 

Before we start going for a lower noise frequency to put the intermod products in band, remember, the noise level drops as you move down in frequency. Lower the level of one intermod source frequency and the products lower too, only faster because its a non-linear system.

 

Also, if broad spectrum noise is one component, then the intermod product will also be broad spectrum noise.

 

I guess before we continue the discussion somebody should specify something about the noise.

Hold on, why does the other frequency have to be in the audible band? Why couldn't it be 45 kHz, for example? In that case, the difference would be 5 kHz, and the harmonics of this frequency would also be affected. I also don't understand what you mean when you say noise is greater at higher frequencies. 

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › What makes piano sound so hard to reproduce?