Article: Realism vs Accuracy For Audiophiles | Part 2: The Real Sounds Of Live Music

[ARQuint]

 

Part 2 is, if anything, more spectacular than Part 1. It illuminates the recording arts, the art of listening, and to me—the meaning of the audiophile pursuit.

 

It got me thinking again about a project I started a few years back, and have mentioned here once before. In 1998, the Chicago-based rare instrument dealer Bein & Fushi was responsible for the project of gathering in one place 30 of the world's most prized violins—15 Stradivaris and 15 Guarnieris. The instruments were photographed and each one played by the American virtuoso Elmar Oliveira. He was recorded by Mark Levinson (as well as by Jerry Bruck, though I don't know if that version ever saw the light of day.) The result was a beautiful 13" x 13" book that came with three CDs. There weren't very many copies produced but I reviewed it for TAS over 20 years ago and was permitted to keep mine. The assemblage of violins, by the way, was valued at roughly $100 million; I can only imagine what the value of the 30 instruments would be now, and what it cost to insure the undertaking back in 1998.

 

https://www.amazon.com/Miracle-Makers-Stradivari-Guarneri-Oliveira/dp/0966174208

 

Oliveira was recorded in a broad range of repertoire, but one of the three CDs was devoted to him playing the opening minute of the Sibelius violin concerto (unaccompanied, of course) on all 30 instruments. As my system improved, I became increasingly interested in whether listeners of various stripes could learn to tell the difference between a Strad and a Guarneri—and how they would do it. So I made a FLAC file of 12 of the 30 iterations of the Sibelius excerpt and—after letting the test subjects listen to several examples "sighted"— arranged them in random order and asked them to identify the instrument they were hearing as either an S or a G.

 

My "n" was/is small, but most test subjects to date haven't done better than guessing. There were three exceptions, though. Two were violinists—a middle-aged female and another woman who was in her 90s. The third was a good friend who would be proud to be called an audiophile, though he certainly knows a lot about music and we attend classical concerts together. He got a perfect score—12/12 instruments correctly identified. No one else came close.

 

I asked the audiophile friend then, and again yesterday, how he did it. Did he have a set of descriptors that he used to characterize each example, devised from listening to the "sighted" examples. I referred him to Bluesman's article and he answered me a few hours ago:

 

"… the heuristic I used was probably not very precisely defined (“list-y”) or calibrated. I’d say it was an attempt to infer from the known references a timbral signature, a kind of holistic “sound.” The best way I could describe it would be that the Strads seemed to speak quicker, to project power more quickly on attack, and the Guarneris seemed to have a mellower/warmer tone and to reach full power fractionally more gradually. After reading Bluesman’s piece and the comments, I’m inclined to hypothesize that both qualities might have to do with differences in the relative strengths/shapes of the fundamental and harmonic series between instruments from each maker."

 

I feel that this helps vindicate—if any vindication was necessary—the manner in which we address ourselves to recordings and to gear. It's how we characterize the success of engineers and manufacturers, beyond a strictly emotional response to what we're hearing. There's a language, a lexicon, that may sound unnecessarily effete to the uninitiated and there is, of course, no requirement that you be able to explain why a given recording or piece of audio equipment seems "realistic" and gets your juices flowing like the real thing does. But many enthusiasts find this an essential part of what we do as audiophiles.

 

 

 

[ARQuint]

5 hours ago, Jud said:

 

Were there also people who did much worse than guessing?

 

There were.

 

I found 13 answer sheets, and these were the results:

 3/12 correct

 4/12

 5/12

 6/12

 6/12

 6/12

 7/12

 7/12

 8/12

 8/12

 9/12       (violinist)

 9/12       (violinist)

 12/12     (audiophile)

 

I don't pretend that this goes much beyond "anecdotal" but one other part of the experiment was interesting to me. The subjects were actually presented with two differently randomized series of files, the first encoded as 16-bit FLAC and the second as 145 kbps MP3. My "star"—the audiophile who got 12/12 correct with FLAC—got just 7/12 right with the MP3 trial.

 

Maybe you remember, Jud, I'd mentioned the idea of an online version of the trial. Originally, my thought was to do this via the TAS website but as the site is no longer "interactive" (a good thing, IMO, given the frequency of childish and totally OT posts) maybe we could do it here, if Chris was on board. Any interest?

 

Andy

[ARQuint]

11 hours ago, Jud said:

 

One other question before deciding whether I'd be up for listening: To the extent possible, are the selections volume matched?

 

Interesting question. The 30 instruments were recorded one after another over 2 days, using the same recording set-up. When I measure the average level from the listening position for each one minute Sibelius excerpt, mostly they are pretty close—but some are a few dB different. This is to be expected.  Oliveira is playing 30 different instruments and some are "louder" than others: the "size" of the sound is an important characteristic of each individual violin—and why one player might decide to pay 5 million for one Stradivarius but not for another. I think it would be a big mistake to adjust the volume of each violin to the same level - that would be distorting an inherent quality of each instrument.

[ARQuint]

21 hours ago, CG said:

So, while an appreciation and understanding of why different instruments sound different based on who is playing may be very helpful for some - maybe all - people, I'm not certain that this understanding will help everybody get more out of the home audio systems.  I'm not saying it's not worth trying or that it's not interesting, but I suspect that a lot of people look for different cues in sound in their quest for accuracy and realism that are beyond the details behind the musical performance.  That part's a whole 'nuther story.

 

I disagree. I know I've said this before, here and elsewhere, but I'll say it again: The audiophile pursuit is "about" that point of intersection between art and technology. I think that's a key take-away of bluesman's series of articles, no matter what your taste in music is.

 

To elaborate on an example that bluesman mentioned, Berlioz specified that a cornet play the obbligato in the "Valse" movement of Symphonie fantastique, as did Stravinsky for one of the two brass in the ensemble for L'histoire du soldat. Assuming that the conductor has followed the composer's instructions and the player is using the correct instrument (and the recording has been competently engineered and produced) if it sounds like a trumpet, something's wrong with your system. It really is as important to be able to read a score as it is an electronics schematicto fully assess (and enjoy) good audio gear and recordings.

 

So I'd like to proceed with continuing the experiment I described above. With Chris's help, with a new thread, I'll post three folders of files for interested AS members to listen to—all containing examples of the one-minute Sibelius excerpt played by the violinist Elmar Oliveira. One folder will contain 12 examples, unidentified, played on 12 different Strads or Guarneris in 16/44 resolution; the second will contain the same 12 examples (in a different order) converted to 145 kbps MP3. The third folder will offer six examples of identified Stradivarius and Guarneri instruments—different ones than the ones presented as "unknowns."

 

Participants can do their own research, but I'll include some expert description of the characteristic sound of the two "brands" of old Italian violins. I'll ask participants to identify each track as "S" or "G" and send me their results as a PM. We'll keep it open for 2 weeks and I'll then present the data. My brother's an academic mathematician and said he'll help me with the basic statistical analysis, though I'm certain that plenty of AS members could assist here, as well.

 

Hope this get this going by tomorrow morning.

 

Andy

[ARQuint]

1 hour ago, Jud said:

 

Yes, it would (or it might simply be how he played the instrument on the day - he's not a robot).

 

But loudness variation (even an average of 1dB) creates a big problem. There was a listening test here years ago where a file that had its average level increased 1dB was the runaway winner. And if you look at the stories about blind competitions where carbon fiber violins beat Stradivarii, guess what you find? Loudest wins.

 

So that's a problem in a blind test.  I don't know what a Strad or a Guarnerius sounds like, so if asked to distinguish, the first thing my ear-brain will fasten on is loudness differences.

 

I suppose we could try with equalized and non-equalized volume and see what the results are?

 

I think it would be entirely up to the individual participant whether or not to volume match. Since we're not looking for a "winner" here but differences, leaving those few files that are a little louder or softer as is might actually help with identification.