[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: GG:has the electronic future arrived?

While rereading our old messages on archive I came upon a particularly
thought-provoking one from Bradley (quoted below).

Regarding computer improvisation, I wonder if any of our fellow
f-minorians have heard of the music software "Band-In-A-Box".  One of its
capabilities is to "improvise" according to a preset sequence of chord
changes.  The user can enter an arbitrary chord sequence, and can even
choose the style (among dozens of them) of improvisation.  I was swept off
my feet when I heard it produce a stream of slick guitar licks a la Pat

And this software is pretty affordable at around 100USD.  I suspect
there's some "neural net" working behind the scenes, with each
improvisation "style" being a set of pre-trained net nodes.  But just on
the face of it, it's really amazing!  (Note: and I'm not talking about
state-of-the-art technologies either, as it's available over a year ago.)


On Fri, 25 Feb 2000 15:27:24 -0500, "Bradley Lehman" <bpl@umich.edu>

 |(Yesterday's NYTimes article is now reproduced on
 |http://www.superconductor.com )
 |About SuperConductor's "performance" I wrote yesterday:
 |>Mozart's Piano Concerto K453, 3rd Movement.  Also not too bad, though again
 |>the large paragraphs don't breathe.  I miss playfulness and whimsy.  But
 |>biggest oversight is a fundamental element of musical grammar.  The program
 |>clobbers the resolution note of every appoggiatura (and this movement has
 |>hundreds of them at the ends of phrases) rather than resolving quietly.
 |>This error is as musically serious and obtrusive as using the word "the" at
 |>the end of a sentence, or putTING acCENTS on the wrong syllaBLE.  You'd
 |>think that given all the trouble they've gone to in this program,
 |>with regard to dynamics, they'd teach it how to recognize this.
 |Thinking some more about this and listening to a good performance of the
 |Mozart concerto K453 (a live recording with Robert Levin at the piano),
 |another severely lacking element occurred to me.  It can be described very
 |simply: SuperConductor doesn't improvise.
 |Even in the most severely detailed and thoroughly notated classical pieces,
 |in which one is "not supposed to" improvise any new notes, there is still an
 |element of improvisation at play all the time.  Every performance is
 |different as the performer reacts to the instrument, the room, any listeners
 |present, her personal feelings at the moment, and what she had for lunch.
 |Tempos, accents, phrasing, articulation, and many other elements are subject
 |to change according to the needs of the moment.  (If the listeners aren't
 |paying attention, make subtle changes to coax their attention back....)
 |This principle should be familiar to anyone who has ever had a music lesson.
 |The teacher or coach suggests, "Try it again and pay attention to X," or
 |"Give it more Z."  The student mentally focuses on X, or adjusts the Z
 |parameter in the brain, and plays the piece differently.  If the student is
 |especially responsive and skilled, the piece comes out sounding *very*
 |different at only a small suggestion.  The score didn't change, but the
 |music certainly did.
 |Musicians who no longer take lessons have trained themselves to give
 |themselves such suggestions to get the type of performance they want at any
 |given moment.  For example, at a recording session a musician plays the same
 |piece several times with different emphases, and then later chooses the take
 |that was most successful musically (reacting this time as a listener, not as
 |a player).  At a live performance the musician chooses emphases depending on
 |what that particular situation seems to need, and depending on an
 |imaginative assessment of what the listeners are receiving.  Recall GG's
 |flight from concerts to recordings: he was weary of trying to project the
 |music with "party tricks" so it sounds convincing at multiple points in the
 |room.  That's a very real problem for musicians, actors, a public speaker,
 |anyone on a live stage.
 |Preparation for a performance is a matter of preparing a whole range of
 |parameters that can be tweaked as needed at the moment, responding to what's
 |going on.  Sometimes it's playing all the same notes but in a different
 |manner; sometimes it's playing some different notes.  It's *not* going to a
 |practice room, getting something absolutely perfect "as the composer
 |intended it", and reproducing exactly the same thing in performance.  An
 |overrehearsed performance is rarely an interesting performance.
 |The brinkmanship of not knowing exactly how it's going to come out is part
 |of the joy of playing music.  Every time it's different.  If the composition
 |is rich enough, it's still interesting after hundreds of performances.  If
 |it's going very well, the audience is swept up in the sense of adventure and
 |helps to *cause* some of the on-edge improvisation.
 |One attempts to simulate this in recordings.  Even though it will sound
 |exactly the same on replay, there has to be enough human irrationality built
 |into the performance to make it feel fresh and repeatably rich-textured.  If
 |one merely gets all the notes right in an objective manner, the result
 |sounds antiseptic and doesn't stand up well to repeated listening.  The same
 |result happens if one over-edits a recording...it doesn't feel real anymore.
 |It might as well be a machine playing.  (The reason we keep listening to GG
 |recordings is that he was a master at providing a rich texture and a sense
 |of fresh adventure: there's something to hear and pay attention to every
 |In a piece such as this Mozart concerto, where the performer *is* supposed
 |to improvise new notes as well as playing the written notes, SuperConductor
 |has no imagination.  It just goes through reproducing everything it sees
 |according to its recipe.  Such a performance doesn't repay close listening;
 |it only superficially sounds OK.  The program also does not recognize that
 |the exact same notes at different points in a composition have different
 |purposes (for example, the way the aria of the Goldberg Variations sounds
 |different at the beginning and end of a performance).  A human performer,
 |having lived and aged through the course of the performance, does play
 |identical notes differently in repeats.  The audience has lived and aged,
 |too, and needs something different the second time.
 |It's the irrationality that makes good performances breathe like living
 |things.  It's a carefully controlled willingness to let tiny random things
 |happen in the moment, the spark of creativity.  The good musician has a
 |whole range of "correct" interpretive parameters to choose from as needed,
 |not just one.  It's a fantastic experience when a performer allows herself
 |to discover new things about a composition *during* a performance, as it
 |unfolds: a sensitive listener can sometimes sense when this is happening.
 |Computers don't have imagination or feelings, they don't sense meaning, and
 |they don't learn from experience.  That's why computer programs will never
 |be able to give performances that sound like live musicians, at least until
 |they are able to adapt from moment to moment to the situation.  Somebody has
 |to build an artificial flexibility and randomness deeply into the program,
 |allowing them to simulate human irrationality and adaptability.  Then they
 |will sound perhaps as good as boring uninspired musicians.
 |It's still not what the composers intended, unless the composers were
 |writing explicitly for mechanical reproduction.  (There *does* exist good
 |music of such a type, for example Conlon Nancarrow's studies for player
 |piano.  Nancarrow makes mechanical precision a virtue rather than a
 |liability.  Let the machine play polyphonic rhythms of 3 against 7 against
 |11 while accelerating at a rate of 2% per minute.  That's what machines are
 |good at.)
 |Bradley Lehman, http://www-personal.umich.edu/~bpl
 |Dayton VA