Re: More about the 20 pieces chosen for adaptive retuning

[mclaren wrote:]
>I have no knowledge of whether MIDI files exist for the 20
>compositions I've chosen, and no interest in whether they do. If we
>restrict ourselves to pieces for which MIDI files are available on
>the web, that's tantamount to severe observational selection. Only
>the simplest and least rhythmically and polyphonically and
>charomatically complex pieces of music are likely to be available on
>the web. At a guess, there are probably 8 bezillion versions of "Fuer
>Elise" and zero versions of the "Grosse Fugue" available on the web.

>Well, that's not my concern. You asked me to pick pieces to
>retune. I picked 'em.

Perhaps you missed my post of July 20, in which I said:

[JdL:]
>>Please let me know if you have any thoughts on a piece for the blind
>>challenge, and I'll try to get it posted this weekend. Natch, it has
>>to be something that I can get ahold of an actual GM sequence of, and
>>that I agree is appropriate (i.e., no serial pieces ;-> ).

[mclaren:]
>I don't have the time to produce MIDI files of these pieces of
>music. So if you want me to spend weeks or months of my precious time
>generateing MIDI files to test your claims which have already been
>knocked down by a significant body of experimental psychoacoustic
>evidence...ain't gonna happen. The burden of work must go to the
>person who desires to prove hi/r claims are accurate in the face of
>large amounts of evidence to the contrary.

Perhaps I have not made myself clear. First, the only claim I have
ever made with certainty is that _I_ like my own work. We have, of
course, had many posts on this very list from others who prefer my
treatments of their own and past works - I can provide exact message
citations if you've missed them. There have also been posts from
people who _don't_ like my treatments. I think the idea of a blind
test is interesting, but I don't feel a need to move heaven and earth
to meet someone else's idea of how it should be done. I am content to
please myself and let history determine how well my ideas are accepted
by others. OK?

I have offered to take the time to tune any sequence I'm able to that
you or anybody might find. What I get from you is one demand after
another. Quite frankly, I feel that if I jumped those hurdles, you'd
find some others to throw up. Please forgive me if I lose interest in
trying to satisfy you.

>You asked "Why not just play all of each piece?" or words to that
>effect.

>The reson for not playing all of each piece in both versions is
>that this will give too many cues to listeners with a bias. For
>example, a diatonic solo melodic line rising from c to F will
>instantly give away the fact that it has been retuned to 5-limit JI
>because of the melodic unevenness twixt C and E and F. Someone who
>is biased in favor of 5-limit JI or biased against 5-limit JI will
>pick up these differences by A-B'ing between the various versions of
>complete piecces of music.

You seem obsessed with the idea that people will "cheat" because they
are "biased".

>To avoid the possibility that listeners will systametically do
>A-B comparisons between identical sections of the same differently
>tuned composition, the sections presented from the same composition
>should not overlap.

I disagree. I take it as given that people are of good will, have no
particular axe to grind other than the promotion of beauty, and can
be trusted with a complete A/B comparison.

>Incidentally, if all this sounds lik e alot of work, you're
>right. Doing a serious double-blind trial involves tons of work.
>That's why people who claim to have conducted "listening tests"
>usually have not done such work, and consequently their claimed
>results don't mean much.

>Inviting a couple of friends over for beer and a listening
>session to various versions of "Feur Elise" from a MIDI file pulled
>down off the net is not a serious test, and no one can take the
>results of such a "listening test" seriously.

Fine, then don't! You seem quite certain you don't/won't like what I
do; that's all that's really necessary for you, right? If some others
of us "drink beer" and enjoy the retuned music, that's no skin off your
nose.

>The alternative -- namely, choosing a variety of
>compositions, subselecting a set at random, creating hte MIDI files,
>producing 2 versions of the audio files, then encoding them into MP3
>and subsequently getting a computer to randomly rename and scramble
>the files, and finally uploaing 'em and analyzing the listener
>results with ANOVA and chi-square techniques -- involves tons of work.
>Well, that's the experimental method. It takes a whole lot of work to
>remove bias and make sure that apparent differences in results are
>not coming from the tester's bias or the listener playing A-B
>listener comparison games or otherwise cheating, or from stiatistical
>artifacts, or from other sources of error.

>Welcome to the wonderful world of the objective double-
>blind experimental method. Designing a good experiment is a subject
>grad students take courses in. People can take years to design an
>conduct a solid reliable experiment.

>It's not trivial. It takes a lot of work. If you're not
>prepared to put in that kind of work, you may want to back off on
>your claims.

Exactly what are you claiming I've claimed? Of the two of us, _you_
are the one who has made categorical claims (that 12-tET is universally
preferred). I know very well that this claim of _yours_ is false, at
least for many people listening to my work, but I have no particular
need to prove that you're wrong. No amount of tests could convince you
otherwise in any case, given how "certain" you are. Just believe
whatever you want to!

>As for entering the MIDI files, doesn't there exist
>software today which can scan in scores and produce MIDI files?
>MIDISCAN or something like that?

>I can provide scores for each and every one of these
>pieces (most of 'em are for the piano and use only 2 staves), so if
>you can find someone with MIDISCAN (or whatever the software's
>called), creating MIDI files should not be a problem.

I don't know what exists in that regard. There are thousands of GM
sequences on prs.net, a lot more than "Feur Elise". Plenty enough, I
would think, to satisfy any reasonable desire for a good mix for
testing.

JdL

John A. deLaubenfels wrote:
>[mclaren wrote:]
>>The reson for not playing all of each piece in both versions is
>>that this will give too many cues to listeners with a bias. For
>>example, a diatonic solo melodic line rising from c to F will
>>instantly give away the fact that it has been retuned to 5-limit JI
>>because of the melodic unevenness twixt C and E and F. Someone who
>>is biased in favor of 5-limit JI or biased against 5-limit JI will
>>pick up these differences by A-B'ing between the various versions
>>of complete piecces of music.
>
>You seem obsessed with the idea that people will "cheat" because they
>are "biased".

Brian is right to avoid bias here, but doesn't seem to understand
what adaptive tuning is, confusing it with static JI in this example.
To be fair, he was only giving an example, but has anyone seen
anything to suggest he's including adaptive JI when he says that
'5-limit JI sounds out of tune'? Brian, John's '5-limit JI' isn't JI
at all; it's dynamically generated temperament. Have you heard it?

>>To avoid the possibility that listeners will systametically do
>>A-B comparisons between identical sections of the same differently
>>tuned composition, the sections presented from the same composition
>>should not overlap.
>
>I disagree. I take it as given that people are of good will, have no
>particular axe to grind other than the promotion of beauty, and can
>be trusted with a complete A/B comparison.

I don't know, John. Brian has a point here.

>Fine, then don't! You seem quite certain you don't/won't like what
>I do; that's all that's really necessary for you, right? If some
>others of us "drink beer" and enjoy the retuned music, that's no
>skin off your nose.

Careful, John. Brian said you said that listeners prefer 5-limit JI.
Instead of denying it, you accepted it as a challenge and even
created a list for it. Brian has outlined what he feels would make
an acceptable test. If you do it and he still isn't satisfied, he
won't have a leg to stand on. Besides, it's not about Brian... it
might be worth knowing.

>>As for entering the MIDI files, doesn't there exist software today
>>which can scan in scores and produce MIDI files? MIDISCAN or
>>something like that?

There are several; MIDISCAN was the first. I believe it is included
with recent versions of Finale, which I have. I've never tried it,
but I'm told it makes a lot of mistakes (surprise, surprise...).

>I don't know what exists in that regard. There are thousands of GM
>sequences on prs.net, a lot more than "Feur Elise". Plenty enough,
>I would think, to satisfy any reasonable desire for a good mix for
>testing.

John's right, Brian. The experiment is at more risk from any one
person selecting 20 pieces than from limiting that selection to GM
files available on the web. The last time I was on prs, they had
no sequence of the Grosse Fugue, but they did have many other
movements from the late string quartets and almost everything Bach
ever wrote. You're right that the burden of work should be on John,
but if you can't provide sequences or pick from available GM files,
you're being unreasonable.

-Carl

[mclaren wrote:]
>>>The reson for not playing all of each piece in both versions is
>>>that this will give too many cues to listeners with a bias. For
>>>example, a diatonic solo melodic line rising from c to F will
>>>instantly give away the fact that it has been retuned to 5-limit JI
>>>because of the melodic unevenness twixt C and E and F. Someone who
>>>is biased in favor of 5-limit JI or biased against 5-limit JI will
>>>pick up these differences by A-B'ing between the various versions
>>>of complete piecces of music.

[I wrote:]
>>You seem obsessed with the idea that people will "cheat" because they
>>are "biased".

[Carl wrote:]
>Brian is right to avoid bias here, but doesn't seem to understand
>what adaptive tuning is, confusing it with static JI in this example.
>To be fair, he was only giving an example, but has anyone seen
>anything to suggest he's including adaptive JI when he says that
>'5-limit JI sounds out of tune'? Brian, John's '5-limit JI' isn't JI
>at all; it's dynamically generated temperament. Have you heard it?

[mclaren:]
>>>To avoid the possibility that listeners will systametically do
>>>A-B comparisons between identical sections of the same differently
>>>tuned composition, the sections presented from the same composition
>>>should not overlap.

[JdL:]
>>I disagree. I take it as given that people are of good will, have no
>>particular axe to grind other than the promotion of beauty, and can
>>be trusted with a complete A/B comparison.

[Carl:]
>I don't know, John. Brian has a point here.

Two versions of each piece, each extracted to tiny pieces chosen
"randomly"? That sounds to me like a _much_ better way to introduce
bias than anything mclaren imagines _I_ could do to influence the
result. I would prefer to post the entire MIDI sequences for those who
would like to hear them, then let anybody who wants to dissect the
works. There could be separate polls, and the dissected pieces could
be given uncorrelated names. This would also provide an interesting
cross-correlation, which could potentially call my own claims into
question (mclaren, you should like this!).

[JdL:]
>>Fine, then don't! You seem quite certain you don't/won't like what
>>I do; that's all that's really necessary for you, right? If some
>>others of us "drink beer" and enjoy the retuned music, that's no
>>skin off your nose.

[Carl:]
>Careful, John. Brian said you said that listeners prefer 5-limit JI.
>Instead of denying it, you accepted it as a challenge and even
>created a list for it. Brian has outlined what he feels would make
>an acceptable test. If you do it and he still isn't satisfied, he
>won't have a leg to stand on. Besides, it's not about Brian... it
>might be worth knowing.

You are a bit confused, Carl. I accepted a challenge for _one_ piece
chosen by mclaren. I'll go so far as to tune 10 pieces, if someone
picks actual available sequences. That has nothing to do with the
tuning-challenges list, which was formed long before this particular
issue came up.

[mclaren:]
>>>As for entering the MIDI files, doesn't there exist software today
>>>which can scan in scores and produce MIDI files? MIDISCAN or
>>>something like that?

[Carl:]
>There are several; MIDISCAN was the first. I believe it is included
>with recent versions of Finale, which I have. I've never tried it,
>but I'm told it makes a lot of mistakes (surprise, surprise...).

[JdL:]
>>I don't know what exists in that regard. There are thousands of GM
>>sequences on prs.net, a lot more than "Feur Elise". Plenty enough,
>>I would think, to satisfy any reasonable desire for a good mix for
>>testing.

[Carl:]
>John's right, Brian. The experiment is at more risk from any one
>person selecting 20 pieces than from limiting that selection to GM
>files available on the web. The last time I was on prs, they had
>no sequence of the Grosse Fugue, but they did have many other
>movements from the late string quartets and almost everything Bach
>ever wrote. You're right that the burden of work should be on John,
>but if you can't provide sequences or pick from available GM files,
>you're being unreasonable.

How about it, mclaren? What is your real goal here? To put my methods
to a real-life test, or to create a script in which you did not receive
enough cooperation, and therefore (exactly as you cleverly predicted)
the test doesn't take place? I don't intent to lose sleep either way:
as I've said before, the worth or lack of worth of my work is not
dependent upon your approval, or lack thereof.

JdL

Brian wrote:

> At a guess, there are probably 8 bezillion versions of "Fuer
> Elise" and zero versions of the "Grosse Fugue" available on the web.

<http://www.netcomuk.co.uk/~ar_wood/GrFuge.html>
<http://member.nifty.ne.jp/continuo/Midi/Stg_fuge/Stgf-e.htm#MIDIDATA>

Graham

--- In crazy_music@y..., xed@e... wrote:
> FROM: mclaren
> TO: new practical microtonality list
> SUBJECT: more about the 20 pieces chosen for adaptive retuning
>
> The reson for not playing all of each piece in both versions
is
> that this will give too many cues to listeners with a bias. For
> example, a diatonic solo melodic line rising from c to F will
> instantly give away the fact that it has been retuned to 5-limit JI
> because of the melodic unevenness twixt C and E and F.

Let's put aside the fact that a diatonic solo melodic line rising
from C to F is already such a short snippet of music that it's far
too short to demostrate many of the capabilities of John's adaptive
tuning program -- so saying it's too long because it gives away too
many clues is dooming this entire experiment from the start.

Let's focus on the statement itself.

Brian (I'm speaking in the third person because that's how he does
it) clearly doesn't know the difference between strict JI and
adaptive JI (let alone adaptive tuning).

In strict JI, the whole step between C and D and the whole step
between D and E will usually be audibly unequal -- one of the whole
steps being 9:8 and the other being 10:9. This difference is about 21
1/2 cents.

In adaptive JI, in its raw form, the sizes of these steps will
usually be much closer to equal. In Vicentino's second tuning of
1555, for example, they will typically be only 5 1/2 cents different.
In John's program the reason is that most pieces of music involve
many of the pitches in many harmonic relationships, and John's
program finds a COFT -- a fixed temperament -- that best "averages
out" these harmonic desiderata. This harmonic "averaging out" usually
results in a melodic "averaging out" of the whole steps in the COFT
so that each can stand for both 9:8 and 10:9 in the lattice. Finally,
the process of finding the final, fully adaptive tuning involves
grounding to COFT and, though it may tend to introduce larger
inequalities between step sizes, these inequalities are still usually
much smaller than those in strict JI.

Furthermore, John deL. has recently introduced an option in his
program that, if activated, tends to equalize the melodic major
seconds, and also to equalize the melodic minor seconds.

By the way, my personal view is that Brian is making this whole
subject (even though he doesn't understand what the program does) 100
times more complex than necessary.

Let's just focus on the psychoacoustic aspect of JI versus non-JI
harmonies, since that's what Brian focused on, and let's look at one
of the studies he brought up.

Matthews and Robinson found that some listeners -- call them "pure"
listeners -- dislike deviation from just ratios, whether that's
because of the beating (the most obvious explanation) or for some
more mysterious reason. Other listeners like the deviation (if it's
not too large) and presumably the beating that's associated with it.
So clearly John's program is going to "improve" pieces from the point
of view of the "pure" listeners, and not from the point of view of
the "rich" listeners. John, and Carl Lumma, are "pure" listeners. So
they like the results of the program. Others won't. Most of us
(Joseph Pehrson, for example), can clearly hear that John's program
is making the harmonies more "pure", or slowing down the
beating . . . without necessarily having to make an aesthetic
decision about whether that's an improvement or not. So to most of
us, the program "works" -- it does what it was made to do. A few
others hear no difference.

That's all there is to it (ignoring the issue of comma shifts, which
was a major obstacle in the way of making musically viable "JI"
renditions in past centuries, but John's explicitly reduces such
shifts to normally inaudible intervals -- thanks in part to my
repeatedly hearing such shifts in earlier versions of the program).
What's a double-blind "preference" experiment going to prove? Some
people will prefer the reduction of beating, some people won't, and
still others won't hear the difference. So the results of John's
program will be "preferred" by the former group but not the other two
groups. But we already know that!

I wrote,

> Matthews and Robinson

Oops -- I meant Mathews and Roberts.