Jump to content

  • Log In with Google      Sign In   
  • Create Account


A lot of generative music


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
19 replies to this topic

#1 pernyblom   Members   -  Reputation: 194

Like
1Likes
Like

Posted 05 September 2012 - 08:48 AM

Hello!

I have improved my generative music system and added some proper voice leading and structure.

EDIT: ...Another Youtube video with some filter effects

EDIT: ...Youtube video playlist with some handpicked songs

Here are 40 songs that the system generated with a single seed as input:

1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40

Yes, they sound a bit similar sometimes and the mixing is sometimes terrible (the melody drowns in other stuff) but it is still kind of cool I think Posted Image

Edited by pernyblom, 18 October 2012 - 04:01 PM.


Sponsor:

#2 bschmidt1962   Crossbones+   -  Reputation: 1699

Like
3Likes
Like

Posted 05 September 2012 - 10:17 AM

Pernyblom,
Interesting. If this is all generated, then your system has some very cool features! I agree that "it is ... kind of cool!" :).
Do you have any links to papers that discuss what you're doing?

That said, I have a suggestion.
Stop using General MIDI sounds. In fact, never use GM again :). To be brutally frank, the quality of the sound samples in your demos are so poor that your listeners will be completely put off by the bad production values of the music.

Can you take exactly the same data but use a real sample library (eg East West, etc).

Because as it stands, no matter how good the generative algorithms you have are, your music will be judged as "oh--that sounds like early 1990-era General MIDI music". And no one will want to see past that.

Brian Schmidt
Register now for GameSoundCon 2012, San Francisco
October 24-25
www.GameSoundCon.com

Brian Schmidt

Executive Director, GameSoundCon:

GameSoundCon 2014:October 7-8, Los Angeles, CA

 

Founder, EarGames

Founder, Brian Schmidt Studios, LLC

Music Composition & Sound Design

Audio Technology Consultant


#3 jwezorek   Crossbones+   -  Reputation: 1606

Like
1Likes
Like

Posted 05 September 2012 - 10:38 AM

At work so can't listen now ... but was wondering if your system is still Java-based? I think that's what you said a while ago...

#4 pernyblom   Members   -  Reputation: 194

Like
0Likes
Like

Posted 05 September 2012 - 11:24 AM

<br />In fact, never use GM again :)


Yeah, it doesn't sound great :)
I am thinking about hiring someone who can do this well. Also, I need a good composer that can help me improve the system and choose at least one specific musical style to target.

#5 pernyblom   Members   -  Reputation: 194

Like
0Likes
Like

Posted 05 September 2012 - 11:28 AM

At work so can't listen now ... but was wondering if your system is still Java-based? I think that's what you said a while ago...


No, it is currently all Javascript. This is a completely new system that was originally intented for web client use, and therefore JS was selected.

#6 jwezorek   Crossbones+   -  Reputation: 1606

Like
0Likes
Like

Posted 05 September 2012 - 11:43 AM

What's the difference between this system and your last system?

#7 pernyblom   Members   -  Reputation: 194

Like
1Likes
Like

Posted 05 September 2012 - 11:49 AM

What's the difference between this system and your last system?


This system uses constraint solving/optimization and planning for almost everything which means that it supports proper voice leading, adaptive figuration and "smarter" harmony planning.
The previous system used none of that. It used patterns for everything without any lookahead. The previous system was however reactive and could change the music according to real-time inputs. The current is a lot more static but can support that eventually.

#8 pernyblom   Members   -  Reputation: 194

Like
0Likes
Like

Posted 05 September 2012 - 01:39 PM

<br />Do you have any links to papers that discuss what you're doing?<br />


I don't have any links but it is a mix of several techniques:
* Constraint-based voice leading (four-part)
* Maximum likelihood harmony planning and figuration
* A lot of tricks to divide the large problems into smaller ones

I once read a thesis about using constraint solvers for voice leading, but I decided to use soft constraints and optimization instead since it is more flexible, but also more expensive.

Most of the techniques/constraints are just directly translated from standard harmony books (harmony, voice leading, phrase structure etc.). I used to be an AI researcher so I know a lot of tricks for translating problems into a computer-understandable form :)

Thanks for giving me good advice about the sound and for your time!

#9 bschmidt1962   Crossbones+   -  Reputation: 1699

Like
2Likes
Like

Posted 05 September 2012 - 06:40 PM

Thanks for the info..
Java explains the MIDI implementation..

Can you have it spit out a level 0 midi file? Then you could pass it off to an orchestrator/recorder, though to be fair you would have to tell them not to change the MIDI file in any way.

Have you looked at how musical phrasing would create subtle (or not so subtle) changes in key velocity (affecting both volume and timbre). I think starting to deal intelligently with velocity as you currently do with note value would make even the GM performances sound massively more 'real'.

Brian Schmidt

Executive Director, GameSoundCon:

GameSoundCon 2014:October 7-8, Los Angeles, CA

 

Founder, EarGames

Founder, Brian Schmidt Studios, LLC

Music Composition & Sound Design

Audio Technology Consultant


#10 pernyblom   Members   -  Reputation: 194

Like
0Likes
Like

Posted 05 September 2012 - 11:45 PM

Can you have it spit out a level 0 midi file? Then you could pass it off to an orchestrator/recorder, though to be fair you would have to tell them not to change the MIDI file in any way.

Have you looked at how musical phrasing would create subtle (or not so subtle) changes in key velocity (affecting both volume and timbre). I think starting to deal intelligently with velocity as you currently do with note value would make even the GM performances sound massively more 'real'.


I can output MIDI files. Those MP3s are actually rendered MIDI files where I tried to use a reasonably good soundfont, but as you say, it doesn't sound good :)

When you speak about note velocity, do you mean the strong/weak distribution between beats/subbeats and measures/hypermeasures? I have dealt with that in one of my previous system but haven't implemented it yet in this new one.

Another thing I'm working on is local tempo changes (ritardantos etc.) and also adding random position offsets which can also make it sound more real.

I also forgot to mention that I have been inspired by the material at harmony.org.uk/, the static/dynamic harmony syntax part.


#11 bschmidt1962   Crossbones+   -  Reputation: 1699

Like
3Likes
Like

Posted 06 September 2012 - 09:54 AM

When you speak about note velocity, do you mean the strong/weak distribution between beats/subbeats and measures/hypermeasures? I have dealt with that in one of my previous system but haven't implemented it yet in this new one.


Yes, though I'd also add music phrasing to the list. Almost nothing will make a MIDI score sound more mechanical than having every note have the same velocity. There are natural ways that people play with varying volume (velocity) on both a micro and macro level. I think tackling that problem (maybe through a method similar to how you deal with pitch) could go a long way towards making your demos sound less mechanical..

Keep us posted!
Brian

Brian Schmidt

Executive Director, GameSoundCon:

GameSoundCon 2014:October 7-8, Los Angeles, CA

 

Founder, EarGames

Founder, Brian Schmidt Studios, LLC

Music Composition & Sound Design

Audio Technology Consultant


#12 pernyblom   Members   -  Reputation: 194

Like
1Likes
Like

Posted 11 September 2012 - 06:05 AM

Added: Youtube video playlist with some handpicked songs

Not much new stuff (some velocity transformations) but it has a better presentation and set of songs (audio is also normalized and compressed).

Edited by pernyblom, 11 September 2012 - 06:06 AM.


#13 swiftcoder   Senior Moderators   -  Reputation: 9499

Like
1Likes
Like

Posted 11 September 2012 - 09:15 AM

Added: Youtube video playlist with some handpicked songs

Not much new stuff (some velocity transformations) but it has a better presentation and set of songs (audio is also normalized and compressed).

That's pretty darn good. It's actually pleasant to listen to, which has not been my general experience with generative music :)

Tristam MacDonald - Software Engineer @Amazon - [swiftcoding]


#14 Moritz P.G. Katz   Members   -  Reputation: 1041

Like
2Likes
Like

Posted 11 September 2012 - 01:05 PM

Hello,

I'm not looking forward to the day when we're replaced by algorithms like this - but I'll have to agree, very impressive! :)

Cheers,
Moritz

Check out my Music/Sound Design Reel on moritzpgkatz.de


#15 pernyblom   Members   -  Reputation: 194

Like
1Likes
Like

Posted 18 October 2012 - 03:56 PM

Added some more videos:

... Filter effects

...Tonicization

Edited by pernyblom, 18 October 2012 - 04:00 PM.


#16 Moritz P.G. Katz   Members   -  Reputation: 1041

Like
1Likes
Like

Posted 18 October 2012 - 07:08 PM

Hey, the one with the filter synths is actually pretty sweet.

The arrangement still seems a bit arbitrary at times with the short breakdowns, but it seems like your program actually came up with a neat little melody.
One thing I've noticed is that ending the track doesn't seem to work very well yet - though they do end on the tonic, it sounds like the song's just falling apart somehow or there's some instrument "lagging behind", doing a little ending motive that seems out of place.

Cheers,
Moritz

Check out my Music/Sound Design Reel on moritzpgkatz.de


#17 pernyblom   Members   -  Reputation: 194

Like
2Likes
Like

Posted 19 October 2012 - 01:17 AM

<br />One thing I've noticed is that ending the track doesn't seem to work very well yet - though they do end on the tonic, it sounds like the song's just falling apart somehow or there's some instrument "lagging behind", doing a little ending motive that seems out of place.<br />


I totally agree. The main problem is that I haven't created any special "end motifs" yet for harmony and bass (and melody actually). The current motifs assume that they should connect with something that follows, which sometimes sounds OK when another phrase follows. Like you say, it sounds bad at endings and and also when there is a lot less content that follows. I cheat with the melody motif by always using a single long note at the end of phrases.

I am also working on to improve the combination of instruments and the breakdowns. The module is now much better parameterized so you can for example change only a small part of a song and keep everything else.

Thanks for listening and the feedback. I appreciate it a lot!

Edited by pernyblom, 19 October 2012 - 01:43 AM.


#18 MA-Simon   Members   -  Reputation: 230

Like
1Likes
Like

Posted 20 October 2012 - 05:23 PM

Very Impressive!
Regarding the endings, you could try something like slowing down the last few notes in some sort of ritardando.

#19 pernyblom   Members   -  Reputation: 194

Like
0Likes
Like

Posted 22 October 2012 - 01:33 AM

Very Impressive!
Regarding the endings, you could try something like slowing down the last few notes in some sort of ritardando.

Thanks!
I insert a song end slowdown with a 30% chance at the moment.
I also try to implement a more dynamic tempo model, using this article as inspiration.

There will be a setting for this in the final tool though :)

#20 Ken Zyma   Members   -  Reputation: 101

Like
0Likes
Like

Posted 30 October 2012 - 08:03 PM

Very very cool stuff! I look forward to hearing this project as it progresses further.

-Ken Zyma




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS