Melodeon.net Forums

Please login or register.

Login with username, password and session length
Advanced search  

News:

Welcome to the new melodeon.net forum

Pages: [1]   Go Down

Author Topic: The Machine Folk Session website  (Read 2294 times)

0 Members and 1 Guest are viewing this topic.

boblsturm

  • Good talker
  • **
  • Offline Offline
  • Posts: 57
    • High Noon, GMT
The Machine Folk Session website
« on: July 24, 2018, 03:01:05 PM »

Hello all,

We have just launched a website related to our research on applying computers to creative ends: https://themachinefolksession.org/ This is linked to our online application: https://folkrnn.org. This is a music generation system trained on all 23,000+ tunes from https://thesession.org.

Also see our composition competition: https://folkrnn.org/competition/

And we are running a Machine Folk Tune of the Month: https://themachinefolksession.org/tune-of-the-month/

Let us know what you think. :)
Logged
Bob L. Sturm, Associate Professor, and student of the melodeon

Rog

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2295
  • Repair and tuning in Hants
Re: The Machine Folk Session website
« Reply #1 on: July 25, 2018, 07:43:48 AM »

Great .. looking forward to exploring this.
The midi tune playback audio on this page doesn’t work on a new iPad running iOS 11
https://themachinefolksession.org/tune-of-the-month/
But it does work in a Chrome browser running on latest Windows OS.

How does this site differ from,, say, Folktunefinder, which also displays the dots and plays them for you?
« Last Edit: July 25, 2018, 07:47:06 AM by RogerT »
Logged

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #2 on: July 25, 2018, 11:21:30 AM »

...How does this site differ from,, say, Folktunefinder, which also displays the dots and plays them for you?

If I've understood it correctly, the web application is built on the neural-network-learning program
which 'composes' a new tune every time you kick it into action.

t'other Roger...

Later: Note to Bob Sturm - would it be an idea to incorporate the RNN properties, prime token
values and generation date/time into the ABC code? Probably as comments (%)? Just a thought.
You might also consider cross-posting this to concertina.net?
« Last Edit: July 25, 2018, 12:05:13 PM by lachenal74693 »
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

boblsturm

  • Good talker
  • **
  • Offline Offline
  • Posts: 57
    • High Noon, GMT
Re: The Machine Folk Session website
« Reply #3 on: July 26, 2018, 07:44:47 AM »

If I've understood it correctly, the web application is built on the neural-network-learning program
which 'composes' a new tune every time you kick it into action.
Yep, this is correct. Sorry that the midi playback doesn't work on the ipad. I just tried mine and it is definitely buggy! I think that's a problem with the library we are using.

Later: Note to Bob Sturm - would it be an idea to incorporate the RNN properties, prime token
values and generation date/time into the ABC code? Probably as comments (%)? Just a thought.
You might also consider cross-posting this to concertina.net?
Thanks! We will include that information soon.
Logged
Bob L. Sturm, Associate Professor, and student of the melodeon

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #4 on: July 26, 2018, 02:08:49 PM »

Later: Note to Bob Sturm - would it be an idea to incorporate the RNN properties, prime token
values and generation date/time into the ABC code? Probably as comments (%)? Just a thought.
You might also consider cross-posting this to concertina.net?
Thanks! We will include that information soon.

Ta. Of course, now I look closer, I see that the prime tokens are already there, so it's the RNN
properties, and generation date/time that it might be useful to save as an integral part of the ABC
code.

As I explore a little more, I see there is an 'Archive' button. Does pressing this mean that a copy
is archived in some database which you and your team can monitor? Does not pressing it mean
that the tune is never seen except by the 'composer', ie: the person who hit the 'Compose' button?

Ta.

Roger
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #5 on: July 26, 2018, 02:37:24 PM »

This looks like fun! I have already generated a 4/4 tune which sounds good after inserting a R:Hornpipe,
and a 6/8 tune which is also pretty cool.

First seriously techie question:

I just generated a 2/4 tune:

Code: [Select]
X:2798
M:2/4
K:Cmaj
|:A,>B,|C>A,G,/2A,/2B, |CDCD/2E/2 |FG/2A/2G/2F/2E/2D/2 |
EDC>D |E>DCA,/2B,/2 |CA,A,G, |E/2E/2F/2G/2AG |
E2C3/2:||:D/2 |E/2F/2GAB |c2B/2c/2 B |
AG/2A/2BA |G2F/2E/2D |E/2D/2CD/2E/2F |E/2D/2CDE/2F/2 |
GAGF |E2C:|

Now, this seems to play back as if it had a default note length value of L:1/16. If that
is the case, shouldn't the L:1/16 value be explicitly included in the ABC code? Otherwise
the innocent bystander (ie: me  (:) ) might assume that the default value(*) of L:1/8 is
in force.

Putting it another way - I thought that a bar of 2/4 music had 4 quavers in it (I just looked
at about 10 2/4 tunes). The 2/4 tune above only seems to have 2 quavers per bar?

Whatever, there seems to be some sort of inconsistency/incompleteness here?

Or have I completely misunderstood - I have a feeling I might have got my quavers and
semi-quavers confused...

Roger

(*)I have always understood that the default note length in ABC is L:1/8, which is equivalent
to 1 quaver (1 eighth-note in American terminology). So, if L:1/8, is not explicitly present, it
may be assumed...
« Last Edit: July 26, 2018, 03:14:51 PM by lachenal74693 »
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

Steve_freereeder

  • Content Manager
  • Hero Member
  • ***
  • Offline Offline
  • Posts: 7511
  • MAD is inevitable. Keep Calm and Carry On
    • Lizzie Dripping
Re: The Machine Folk Session website
« Reply #6 on: July 26, 2018, 03:20:22 PM »

I just generated a 2/4 tune:

Code: [Select]
X:2798
M:2/4
K:Cmaj
|:A,>B,|C>A,G,/2A,/2B, |CDCD/2E/2 |FG/2A/2G/2F/2E/2D/2 |
EDC>D |E>DCA,/2B,/2 |CA,A,G, |E/2E/2F/2G/2AG |
E2C3/2:||:D/2 |E/2F/2GAB |c2B/2c/2 B |
AG/2A/2BA |G2F/2E/2D |E/2D/2CD/2E/2F |E/2D/2CDE/2F/2 |
GAGF |E2C:|

Now, this seems to play back as if it had a default note length value of L:1/16. If that
is the case, shouldn't the L:1/16 value be explicitly included in the ABC code? Otherwise
the innocent bystander (ie: me  (:) ) might assume that the default value(*) of L:1/8 is
in force.

Putting it another way - I thought that a bar of 2/4 music had 4 quavers in it (I just looked
at about 10 2/4 tunes). The 2/4 tune above only seems to have 2 quavers per bar?

Whatever, there seems to be some sort of inconsistency/incompleteness here?

Or have I completely misunderstood - I have a feeling I might have got my quavers and
semi-quavers confused...

There might be some hidden bug which has put a default note length of 1/16. So by including an explicit note length field of L:1/8, the note values then look OK when viewed in an ABC browser/printer. See revised code below. I've also rationalised your beaming and line lengths. Finally, I've added 1st and 2nd time bars at the end of the B-music which then preserves the correct number of beats on the repeats.
Interesting little tune....

X:2798
M:2/4
L:1/8
K:Cmaj
|:A,>B,|C>A, G,/2A,/2B, |CD CD/2E/2 |FG/2A/2 G/2F/2E/2D/2 |ED C>D |
E>D CA,/2B,/2 |CA, A,G, |E/2E/2F/2G/2 AG |E2:|
C3/2D/2 |:E/2F/2G AB |c2 B/2c/2B|AG/2A/2 BA |G2F/2E/2D|
E/2D/2C D/2E/2F |E/2D/2C DE/2F/2 |GA GF |1E2 C3/2D/2:|2E2 C:|
« Last Edit: July 26, 2018, 03:33:08 PM by Steve_freereeder »
Logged
Steve
Sheffield, UK.
www.lizziedripping.org.uk

boblsturm

  • Good talker
  • **
  • Offline Offline
  • Posts: 57
    • High Noon, GMT
Re: The Machine Folk Session website
« Reply #7 on: July 26, 2018, 04:30:27 PM »

As I explore a little more, I see there is an 'Archive' button. Does pressing this mean that a copy
is archived in some database which you and your team can monitor? Does not pressing it mean
that the tune is never seen except by the 'composer', ie: the person who hit the 'Compose' button?

Thanks Roger. If you hit "archive", you and your tune will be taken to https://themachinefolksession.org/ where you can submit it as a machine folk session tune!

First seriously techie question:

I just generated a 2/4 tune:

Code: [Select]
X:2798
M:2/4
K:Cmaj
|:A,>B,|C>A,G,/2A,/2B, |CDCD/2E/2 |FG/2A/2G/2F/2E/2D/2 |
EDC>D |E>DCA,/2B,/2 |CA,A,G, |E/2E/2F/2G/2AG |
E2C3/2:||:D/2 |E/2F/2GAB |c2B/2c/2 B |
AG/2A/2BA |G2F/2E/2D |E/2D/2CD/2E/2F |E/2D/2CDE/2F/2 |
GAGF |E2C:|

Now, this seems to play back as if it had a default note length value of L:1/16. If that
is the case, shouldn't the L:1/16 value be explicitly included in the ABC code? Otherwise
the innocent bystander (ie: me  (:) ) might assume that the default value(*) of L:1/8 is
in force.

Putting it another way - I thought that a bar of 2/4 music had 4 quavers in it (I just looked
at about 10 2/4 tunes). The 2/4 tune above only seems to have 2 quavers per bar?

Whatever, there seems to be some sort of inconsistency/incompleteness here?

Yes, the L:1/8 needs to be made explicit in the ABC. It's a problem with the interpreter we are using to render the ABC as staff notation. I haved added it for you: https://themachinefolksession.org/tune/149

Another thing to notice here is that some of the measures are not counted correctly. The system is literally producing the ABC text and will sometimes miscount (like real people do in thesession.org).

I have added a setting of your tune, which I call "Roger's Quavers": https://themachinefolksession.org/tune/147#setting-1
It's a really odd tune because the pickup does not sound like a pickup.
« Last Edit: July 26, 2018, 05:15:34 PM by boblsturm »
Logged
Bob L. Sturm, Associate Professor, and student of the melodeon

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #8 on: July 26, 2018, 06:29:19 PM »

There might be some hidden bug which has put a default note length of 1/16. So by including an explicit note length field of L:1/8, the note values then look OK when viewed in an ABC browser/printer. See revised code below. I've also rationalised your beaming and line lengths. Finally, I've added 1st and 2nd time bars at the end of the B-music which then preserves the correct number of beats on the repeats.
Interesting little tune....

Ta. See Bob Sturm's subsequent post. From what he says, it looks as if there are some glitches in whatever
part of the system is generating (and outputting) the ABC code. When I spotted what seemed to be these glitches
I just posted the query, without bothering about the actual way it was laid out, but you are quite right - it needed
prettying up a little. I like to think that I would have done this, if I hadn't been distracted by the inconsistency (?)
with the L: header, but you have done the job for me - thanks.

It ain't just computer-generated music which has inconsistencies like this. Even with my very sparse musical
knowledge, I fairly frequently spot 'problems' with posted tunes which require (typically) modification of the
endings of the parts to get the beat count correct. Getting sensible line lengths, etc. is something I do all the
time. Whisper it quietly, but sometimes I even alter the tune slightly :o ...

[does that repeat at the end of the B: part go better as: ...|1E2 C3/2D/2:|2E2 C2|] - see what I mean... (:)]

ATEOTD, it don't matter - when you put the score in front of 'real' musicians, they just shrug their shoulders
and play it correctly 'on-the-fly', but for a klutz like me (and for ABC), it's more or less essential that it's 'right'.

I guess any problems with this software will be sorted out eventually? After all, it is a Beta version. As
some-one who used to earn his daily crust in the high-end scientific computing racket, I find it fascinating.
Others may think differently of course...

Roger
« Last Edit: July 26, 2018, 07:01:41 PM by lachenal74693 »
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #9 on: July 26, 2018, 06:48:49 PM »

...I have added a setting of your tune, which I call "Roger's Quavers": https://themachinefolksession.org/tune/147#setting-1
It's a really odd tune because the pickup does not sound like a pickup.
Fame at last! Thank you. Roger.
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

Steve_freereeder

  • Content Manager
  • Hero Member
  • ***
  • Offline Offline
  • Posts: 7511
  • MAD is inevitable. Keep Calm and Carry On
    • Lizzie Dripping
Re: The Machine Folk Session website
« Reply #10 on: July 26, 2018, 07:56:11 PM »

does that repeat at the end of the B: part go better as: ...|1E2 C3/2D/2:|2E2 C2|] - see what I mean... (:)

Actually, the 2nd time bar needs to be only one crotchet long in order to 'pay back' the 'borrowed' pick-up half-bar at the very beginning. It's standard music convention which also enables a repeat back to the very beginning (i.e. when playing the tune multiple times through) to occur without a jerky rhythmic clunk.

So in the context of your tune I would propose that the end of the B-music should read as follows:
...|1 E2 C3/2D/2:|2 E2|]

Hope this helps/make sense...
Logged
Steve
Sheffield, UK.
www.lizziedripping.org.uk

Roger Hare

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 828
  • Urmston, Lancashire, U.K.
Re: The Machine Folk Session website
« Reply #11 on: July 27, 2018, 02:29:18 AM »

...So in the context of your tune I would propose that the end of the B-music should read as follows:
...|1 E2 C3/2D/2:|2 E2|]

Hope this helps/make sense...

It do. I like to think I would have made that adjustment to the music if I were writing it down to be played
through several times, but who knows... See what I mean about 'real' musicians? You are one, I ain't...

Ta.

R.
Logged
For more about Manchester Morris, The Beech Band Folk Club or anything else,  please use the private messaging facility.
My (large) ABC Tune Book is here.

Steve_freereeder

  • Content Manager
  • Hero Member
  • ***
  • Offline Offline
  • Posts: 7511
  • MAD is inevitable. Keep Calm and Carry On
    • Lizzie Dripping
Re: The Machine Folk Session website
« Reply #12 on: July 27, 2018, 07:42:37 AM »

See what I mean about 'real' musicians? You are one, I ain't...
Thanks for the seal of approval, but I always maintain that if you can play an instrument or sing, you can consider yourself a real musician.  (:)
Logged
Steve
Sheffield, UK.
www.lizziedripping.org.uk
Pages: [1]   Go Up
 


Melodeon.net - (c) Theo Gibb; Clive Williams 2010. The access and use of this website and forum featuring these terms and conditions constitutes your acceptance of these terms and conditions.
SimplePortal 2.3.5 © 2008-2012, SimplePortal