Is this targeting people who can already play instruments? I’d be interested in a raw audio based generation where I could interact with an RL agent by giving thumbs up/down to generated chunks. Does it make sense?
The target is more towards people who are interested to experiment on RL or EA in sequential parameters generation.
Instrument playing experience is not a prerequisite at all while some knowledge on synth sequencer or drum machine can be very helpful for understanding the concept.
It seems that your use case can fit this toolkit while you are performing part of the environment to give reward. But I would doubt whether you can be patient enough for so many iterations. :)
My use case would be hooking up the RL agent to something like OpenAI JukeBox, and crowdsourcing thumbs up/down on generated chunks to train the agent.
The most time consuming and tedious part of a music generation engine today is choosing a decent chunk out of (very) many generated samples. If RL can help with that it would be awesome.
Then you give feedback to each bar. Ideally, the NN will adapt to your feedback. For example, you only thumb up to pentatonic notes, the NN should be able to stick to them while occasionally be 'creative'.
This is known as online machine learning, and it has been there quite of some time. JukeBox and RaveForce are just totally different stories.
