Yahoo Tuning Groups Ultimate Backup makemicromusic Audio realization of Bull-'Ut Re Mi Fa Sol La'

Hey,

I resuscitated my old pet-project -- Bull's 'Ut Re MI Fa Sol La' in 19-equal,
and got further along in it tonight...I think in a week I could complete it.

http://www.dividebypi.com/audio/bull_ut_re.ogg

Software and platform info:

*Gentoo Linux
*fluidsynth, using 'Campbell's Harpsichord' soundfont (corrected and retuned
version)
* my own 'et_compose', written in Python, a 'midi file compiler' for writing
in arbitrary n-tets and a healthy alternative to writing Scala *.seq files,
which has the following advantage over Scala's format:
1) I can write polyphonically and *see* the lines over each other, so I know
harmonically and contrapuntally what's going on, and
2) I implemented a 'humanize' function, using Gaussian distribution curves,
that, given the right parameter, will make subtle micro-temporal adjustments
in the output midi file, giving a dynamic and lifelike performance to the
file that would otherwise suffer from being stale and too precise.

The latter I think I used to good effect in the OGG sample above...I tried
about 10 different values until I found one I think is just right for this
performance...just barely audible, but present and interesting. The result
is, I think, when paired with the excellent harpsichord soundfont, nothing
short of ear-candy !!!

Enjoy!
--
Aaron Krister Johnson
http://www.dividebypi.com
http://www.akjmusic.com

🔗Bill Sethares <sethares@...>

Hey Aaron -- nice rendition... keep it coming!

Aaron wrote:

> I implemented a 'humanize' function, using Gaussian
> distribution curves, that, given the right parameter,
> will make subtle micro-temporal adjustments
> in the output midi file, giving a dynamic and
> lifelike performance to the file that would otherwise
> suffer from being stale and too precise.

When I read this I began to wonder why you chose to use
the bell curve rather than some other distribution.
For example, if you chose one that was skewed to either
the left (earlier) or the right (later), then you might
get interesting anticipation or lagging effects... does
anyone know if there is a literature about how musicians
in performance typically deal with the beat -- are deviations
generally unbiased? I would guess that different
performers in different styles tend to make systematic
deviations from the beat...

--Bill Sethares

🔗Aaron K. Johnson <akjmicro@...>

On Friday 27 August 2004 09:51 am, Bill Sethares wrote:
> Hey Aaron -- nice rendition... keep it coming!
>
> Aaron wrote:
> > I implemented a 'humanize' function, using Gaussian
> > distribution curves, that, given the right parameter,
> > will make subtle micro-temporal adjustments
> > in the output midi file, giving a dynamic and
> > lifelike performance to the file that would otherwise
> > suffer from being stale and too precise.
>
> When I read this I began to wonder why you chose to use
> the bell curve rather than some other distribution.
> For example, if you chose one that was skewed to either
> the left (earlier) or the right (later), then you might
> get interesting anticipation or lagging effects... does
> anyone know if there is a literature about how musicians
> in performance typically deal with the beat -- are deviations
> generally unbiased? I would guess that different
> performers in different styles tend to make systematic
> deviations from the beat...

Bill,

Thanks for your reply...2 things:

1) each *voice* has it's own displacement, so that they dont move 'en masse'
from the metronomic position. This was on purpose, so as to implement a sort
of quasi-organic 'flamming' feeling, like one naturally gets in the
microdifferences in ensemble playing, or in finger differences at a keyboard,
etc.

2) *skewing* early or late randomly would tend to make it feel like it was
rushing or dragging over the long haul, which I didn't want. And besides, I
think those things should convey what is going on musically, for instance
delaying slightly a dissonance to a consonance in an appogiatura. So in
answer to your questions about what musicians do, which I know first hand
being a frequent performer and colleague of some great performers, these
draggings (which are probably more common, I suspect, because of rubato) or
rushings happen in response to affective or structural considerations in the
music, at least when the performer is a master, and deliberately doing it.
But, short of writing an advanced context sensitive artificial intelligence
program, or hand entering such values myself, I do believe the Gaussian
distribution curve breathing life into the file is the next best 'automatic',
albeit slightly imperfect, solution. But I have no pretense that it is
solving an *interpretive* problem, only a sonic one, and I think it succeeds
admirably at making the surface of the music more interesting in its
imperfection!

Cheers,

Aaron Krister Johnson
http://www.dividebypi.com
http://www.akjmusic.com

🔗Prent Rodgers <prentrodgers@...>

Aaron,
This realization is great. I've listened to it a few times and get a
kick out of some of the harmonies. There is something really wierd at
the 48 second mark, with a modulation that I don't think the composer
considered. But I like it. The temporal adjustments at 1:55 are great.
It really sounds like someone playing the harpsichord, with the gentle
almost arpegiated chords.

The next step is to take the aspect you said couldn't be done and do
it. That is, "for instance delaying slightly a dissonance to a
consonance in an appogiatura". Why not?

Prent

> > Aaron wrote:
> > > I implemented a 'humanize' function, using Gaussian
> > > distribution curves, that, given the right parameter,
> > > will make subtle micro-temporal adjustments
> And besides, I
> think those things should convey what is going on musically, for
instance
> delaying slightly a dissonance to a consonance in an appogiatura.

🔗Joseph Pehrson <jpehrson@...>

🔗Aaron K. Johnson <akjmicro@...>

!!! Update !!!

I have completed my John Bull 'Ut Re Mi Fa Sol La' realization a couple of
days ago....and am just now posting it. It's nice to be done with a project
that I've been delaying for months!

I left information from my original post a while back (when it was *almost*
done) below for reference.

Here it is (complete version as a .ogg file):
http://www.akjmusic.com/audio/bull_ut_re.ogg

As an incomplete mp3 file:
http://www.akjmusic.com/audio/bull_ut_re.mp3

Enjoy!

Yes, I'm posting it complete only as an .ogg file. This is not to alienate
listeners, lose audience, etc, but because I believe in maximal fidelity in
sound for a compressed format, and I don't believe in sacrificing that to the
degree that I can help it. It is clearer to me than ever after doing
extensive comparisons and research over the last couple of days that .ogg
files are superior to .mp3s, in spite of the arguments for the ubiquitousness
of .mp3. Nevertheless, I give a sample of the last minute and a half or so of
the realization as an mp3; hopefully, if any of you stalwart mp3 users are
intrigued enough, you will join the revolution, and update your listening
situation to include .ogg capability.

For those of you who missed the ogg/mp3 debate, I had a relevant post about
directly comparing their implementation, at least at it occurs on the linux
platform. I included a 'snippet.wav' example, compressed to both mp3 and ogg,
and the results speak for themselves. The article is archived here:

/makemicromusic/topicId_7608.html#7608

Best,
Aaron Krister Johnson
http://www.akjmusic.com
http://www.dividebypi.com

On Thursday 26 August 2004 10:41 pm, Aaron K. Johnson wrote:
> Hey,
>
> I resuscitated my old pet-project -- Bull's 'Ut Re MI Fa Sol La' in
> 19-equal, and got further along in it tonight...I think in a week I could
> complete it.
>
> http://www.dividebypi.com/audio/bull_ut_re.ogg
>
> Software and platform info:
>
> *Gentoo Linux
> *fluidsynth, using 'Campbell's Harpsichord' soundfont (corrected and
> retuned version)
> * my own 'et_compose', written in Python, a 'midi file compiler' for
> writing in arbitrary n-tets and a healthy alternative to writing Scala
> *.seq files, which has the following advantage over Scala's format:
> 1) I can write polyphonically and *see* the lines over each other, so I
> know harmonically and contrapuntally what's going on, and
> 2) I implemented a 'humanize' function, using Gaussian distribution
> curves, that, given the right parameter, will make subtle micro-temporal
> adjustments in the output midi file, giving a dynamic and lifelike
> performance to the file that would otherwise suffer from being stale and
> too precise.
>
> The latter I think I used to good effect in the OGG sample above...I tried
> about 10 different values until I found one I think is just right for this
> performance...just barely audible, but present and interesting. The result
> is, I think, when paired with the excellent harpsichord soundfont, nothing
> short of ear-candy !!!
>
> Enjoy!

🔗Jonathan M. Szanto <JSZANTO@...>

Aaron,

{you wrote...}
>Nevertheless, I give a sample of the last minute and a half or so of the >realization as an mp3; hopefully, if any of you stalwart mp3 users are >intrigued enough, you will join the revolution, and update your listening >situation to include .ogg capability.

Good move (I saw your earlier comments, just got back from a couple days in the Bay area away from computers - YEA!!!). And I'd encourage everyone to check out ogg stuff. If I had the time I'd post links to the encoder I've used, but orch season starts again in about 36 hours and I got a lot of gear to get together...

BTW: is the diff between your ogg file and the wav file greater or lesser than the diff between the ogg and the mp3? (in fidelity, that is)

Cheers,
Jon