spent 1 hour with Rapidcomposer
Posted: Mon Feb 09, 2015 6:50 pm
I quickly went through Rapidcomposer to see whether the cheaper alternative to Cognitone Synfire, was up to any kind of production quality. I mean, the notion itself is a challenge, and already includes a bit of a cheat. Can computers spew out music notation that is useful under the tight supervision of a musician. So in a way, I view it as a supportive tool rather than the program being able to write competent music completely on its own.
Rapidcomposer is the simpler of the two. It has a markov-esque probabilistic chord progression generator, and a phrase generator that is much like a step sequencer that can map polyphonic phrase content to the nearest chord note. (might be able to do more, I haven't gone through everything) So, not too far of from the algorithmic techno generator I made some 12 years ago. The chord generator is actually less intelligent than the one I made, since mine would pick chords taking bars into account, so that it would try to finish off a progression toward the end of the bar, and start off a new bar with a 'start' of a new progression. Rapidcomposer seems to create arbitrary chord change points (in time), and then simply assign chords probabilistically. In my experience, that's too simplistic and doesn't lead to progressions that a composer would pick.
The probablistic step sequencer that rapidcomposer uses I feel is also too simplistic. It blindly uses probability to generate notes, and while it always hits the notes it's supposed to, it leads to a lot of machine gun repeated note triggering. At least mine was based on an oscilating upper cut off point, so the chord comping moved up and down with inversions, even creating pseudo melodies in the process. I think this sort of macro level motion is quite important not just in composing since you use it to create flow and dynamism that spans longer than a 1 measure pattern. Mine step sequencer was even aware of the song structure, so it understood where the build and break points were, to adjust the parameters accordingly.
Overall, I felt Rapidcomposer was much to unstable to be worth the time. It's also functionally too limited to be considered a state of the art algorithmic composition tool. As far as I can see, everything is based on simple probabilistic models... and to my knowledge, AI and computer based decision making should be much more sophisticated. (even as of 12 years ago)
The target demographic is a puzzling one. I looked on their forums, and there seems to be a good amount of users who have a bunch of VSTi and sample libraries (ehem, where they got them from I don't know) who simply use Rapidcomposer as a means to pump notes into them and make sounds. It reminds me of piano library aficionados who only "render" midi files they find floating on the web. Obviously, those are not the developer's intentions.. but I've yet to see anyone who is creating high grade material, incorporating the assistive technologies into a professionally acceptable workflow. There are points where I can see assistive harmonizing and auto notation being helpful, like mocking up strings arrangements. Especially for huge unison staccato ostinato strings that's everyone seems to love for trailor-esque soundtracks. (makes everything sound the same, but.. people want them)
But anyway, I think it's interesting that such piece of software can now exist as commercial software... and so far, both Cognitone and Rapidcomposer are still in business. That's an amazing feat in itself.
Rapidcomposer is the simpler of the two. It has a markov-esque probabilistic chord progression generator, and a phrase generator that is much like a step sequencer that can map polyphonic phrase content to the nearest chord note. (might be able to do more, I haven't gone through everything) So, not too far of from the algorithmic techno generator I made some 12 years ago. The chord generator is actually less intelligent than the one I made, since mine would pick chords taking bars into account, so that it would try to finish off a progression toward the end of the bar, and start off a new bar with a 'start' of a new progression. Rapidcomposer seems to create arbitrary chord change points (in time), and then simply assign chords probabilistically. In my experience, that's too simplistic and doesn't lead to progressions that a composer would pick.
The probablistic step sequencer that rapidcomposer uses I feel is also too simplistic. It blindly uses probability to generate notes, and while it always hits the notes it's supposed to, it leads to a lot of machine gun repeated note triggering. At least mine was based on an oscilating upper cut off point, so the chord comping moved up and down with inversions, even creating pseudo melodies in the process. I think this sort of macro level motion is quite important not just in composing since you use it to create flow and dynamism that spans longer than a 1 measure pattern. Mine step sequencer was even aware of the song structure, so it understood where the build and break points were, to adjust the parameters accordingly.
Overall, I felt Rapidcomposer was much to unstable to be worth the time. It's also functionally too limited to be considered a state of the art algorithmic composition tool. As far as I can see, everything is based on simple probabilistic models... and to my knowledge, AI and computer based decision making should be much more sophisticated. (even as of 12 years ago)
The target demographic is a puzzling one. I looked on their forums, and there seems to be a good amount of users who have a bunch of VSTi and sample libraries (ehem, where they got them from I don't know) who simply use Rapidcomposer as a means to pump notes into them and make sounds. It reminds me of piano library aficionados who only "render" midi files they find floating on the web. Obviously, those are not the developer's intentions.. but I've yet to see anyone who is creating high grade material, incorporating the assistive technologies into a professionally acceptable workflow. There are points where I can see assistive harmonizing and auto notation being helpful, like mocking up strings arrangements. Especially for huge unison staccato ostinato strings that's everyone seems to love for trailor-esque soundtracks. (makes everything sound the same, but.. people want them)
But anyway, I think it's interesting that such piece of software can now exist as commercial software... and so far, both Cognitone and Rapidcomposer are still in business. That's an amazing feat in itself.
