Apologies for my voice sounding a little muffled, I've got a new recording setup and am still figuring out all of my settings. If anyone needs any part of the video explaining, just leave a comment and I'll do my best to reply to all of them :)
Amazing video! I just stumbled upon the first two parts and bought one of your courses on music hackspace a couple days ago so I'm super excited that you released part 3! Also really enjoyed that you used Green Hill Zone. That song is super nostalgic
Hi Samuel, Thanks for these amazing videos! Since you have a different Markov for pitch and duration, does it mean that the learning process is completely separate for these two elements? And do you have any suggestions for connecting these two together? I am trying to have the learning process to be done in a specific order, which is: First having the computer learn the pitch, then the duration based on the pitch, meaning that if the we have a C 1/4 note and D 1/2 note and C 1/4 and then D whole note, then it learns that the C can be only a quarter note and after the C 1/4 there is a 50% chance that the D 1/2 or D whole note appears. I hope my question is somewhat clear!
Hi Hesam, thanks for the question! Yes it does make sense, and currently yes the markov chains are being trained separately. The way they're set up is to only read single numbers in sequence, for what you're talking about you would need some kind of multi-dimensional markov transition matrix that considers pairs of numbers. The ml.markov objects as they currently work won't be able to do that unfortunately. I've built my own bespoke markov chain design in Max for a separate project but it's pretty inefficient right now and can't go above order 2, so I'm not sure it would help either. I'll have a ponder and if I come across anything that might be worth pursuing I'll probably make a video about it eventually.
Thanks for the comment! So in Logic or Ableton you should just be able to click record while playing into a software instrument and it will create a midi file that you can then save, but if you mean in Max itself you can use another 'seq' object. Send it a 'record' message then play MIDI into it from the Markov Chain, then send in a 'stop' message. Finally send in a 'write' message and it will give you the option to save as a MIDI file. Hope that helps!
Apologies for my voice sounding a little muffled, I've got a new recording setup and am still figuring out all of my settings. If anyone needs any part of the video explaining, just leave a comment and I'll do my best to reply to all of them :)
good to see you are back!
Amazing video! I just stumbled upon the first two parts and bought one of your courses on music hackspace a couple days ago so I'm super excited that you released part 3! Also really enjoyed that you used Green Hill Zone. That song is super nostalgic
A big thank you for sharing your knowledge. Big and difficult challenges suddenly seem so easy due to your excellent way of explaining.
thank you! love the addition of live midi input opens up so much possibility!
Thank you very much for this 3 tutorials :)
Thanks so much for this Samuel, I have really enjoyed this series!
Amazing work, very much appreciated!
Hey, it looks really cool, but unfortunately ' MIDI_Length.maxpat' is missing on your dropbox link, where can I find it ?
my bad, is included in the tutorial n 2.
Thanks again for sharing this
Hi Samuel, Thanks for these amazing videos! Since you have a different Markov for pitch and duration, does it mean that the learning process is completely separate for these two elements? And do you have any suggestions for connecting these two together? I am trying to have the learning process to be done in a specific order, which is: First having the computer learn the pitch, then the duration based on the pitch, meaning that if the we have a C 1/4 note and D 1/2 note and C 1/4 and then D whole note, then it learns that the C can be only a quarter note and after the C 1/4 there is a 50% chance that the D 1/2 or D whole note appears.
I hope my question is somewhat clear!
Hi Hesam, thanks for the question! Yes it does make sense, and currently yes the markov chains are being trained separately. The way they're set up is to only read single numbers in sequence, for what you're talking about you would need some kind of multi-dimensional markov transition matrix that considers pairs of numbers. The ml.markov objects as they currently work won't be able to do that unfortunately. I've built my own bespoke markov chain design in Max for a separate project but it's pretty inefficient right now and can't go above order 2, so I'm not sure it would help either. I'll have a ponder and if I come across anything that might be worth pursuing I'll probably make a video about it eventually.
@@Spearced Thank you Samuel!
Great! Thanks for sharing!
And how do you save the generated midi file?
Thanks for the comment! So in Logic or Ableton you should just be able to click record while playing into a software instrument and it will create a midi file that you can then save, but if you mean in Max itself you can use another 'seq' object. Send it a 'record' message then play MIDI into it from the Markov Chain, then send in a 'stop' message. Finally send in a 'write' message and it will give you the option to save as a MIDI file. Hope that helps!
@@Spearced Great! Thank you very much, it's very helpful!
Interesting very
Thanks many