Posted
Climax (aka horizontal Build) and Density (aka vertical Build)
No further explanations needed, I’m pretty sure Andre got it ;-)
Mon, 2018-11-19 - 21:08 Permalink
These parameters would control how many instruments are playing, what they are playing and the rhythm they are playing. So basically, these are generative parameters, rather than transformational ones.
Very cool indeed. I can imagine these in the generative system we are working on ...
Tue, 2018-11-20 - 15:38 Permalink
A very appreciated feature, at least on this side, would be the Synfire ability to orchestrate two or three combinations of instruments -e.g. full orchestra, chamber strings, wind orchestra or anything else- according to the structure provided by the composer. Id est, Synfire should be able to orchestrate a crescendo, an accelerando, a diminuendo and so on for a predefined combo, or at least it should propose some orchestration alternatives.
That would be of great help, you know... :-)
I am 100% ignorant about artificial intelligence, but I suppose that this would mean stuffing Synfire with tons of scores of all genres to make him learn as many timbral combinations as possible...
Tue, 2018-11-20 - 19:58 Permalink
BTW I just listened to an AI generated Nocturne in the style of Chopin.
Nice try, but terrible outcome...
IMO it doesn't make much sense to try to resurrect a composer of the past unless we can rebuild his life, his education and his emotional dynamics. I.e. if we can clone him.
Wed, 2018-11-21 - 20:59 Permalink
AI is great for recognition. For generative purposes it is overestimated. Images and music generated by CNN (convolutional neural networks) or DNN (deep ...) sound like someone is merely dreaming of music, unconscious and foggy fragements pass by, vague associations. It's really cool for a while (and for specific purposes).
But it's cool only because we know how it was made. For any unprepared average listener, there's no composer's intent recognizable, no clear system that grabs you by your ears, slings you around and tells you a story you won't easily forget (the "hook").
Even algorithmic models can generate more contrasted and intentful music at times.
Wed, 2018-11-21 - 21:36 Permalink
AI is great for recognition. For generative purposes it is overestimated. Images and music generated by CNN (convolutional neural networks) or DNN (deep ...) sound like someone is merely dreaming of music, unconscious and foggy fragements pass by, vague associations. It's really cool for a while (and for specific purposes).
I couldn’t have said it better... :-)