1) I think you're right that richness of synthesized sound is a comparatively easy problem. CDs beat vinyl, and eventually we'll have data recordings that even a master musician won't be able to distinguish from live sound.
And I'd agree that musician interface functionality is tougher. You almost need a virtual-reality instrument, some way for a musician to feel the vibrations of the strings in his fingers, or the air pressure in his lungs, the feel of the mouthpiece, etc. Hypothetically this wouldn't have to exactly match any real-world instrument, but it would have to provide the same richness and fidelity of feedback data. Speculating further, such a virtual instrument could be "tuned" to have better interface feedback than the real thing, possibly allowing faster learning or better playing by novices.
I've lost some enthusiasm for genetic-algorithmically evolved music. I figure only someone who's a professional musician already would endure listening to a hundred variations of a lousy song in order to improve it. You might be able to get something like Amazon's book-picker algorithm, though. i.e. if I like songs A1, A2, and A3, I'm likely to enjoy A4 but not B4. And maybe by evolving and abstracting rules for such software, we can find a way to make better musical composer AI.
And I'd agree that musician interface functionality is tougher. You almost need a virtual-reality instrument, some way for a musician to feel the vibrations of the strings in his fingers, or the air pressure in his lungs, the feel of the mouthpiece, etc. Hypothetically this wouldn't have to exactly match any real-world instrument, but it would have to provide the same richness and fidelity of feedback data. Speculating further, such a virtual instrument could be "tuned" to have better interface feedback than the real thing, possibly allowing faster learning or better playing by novices.
I've lost some enthusiasm for genetic-algorithmically evolved music. I figure only someone who's a professional musician already would endure listening to a hundred variations of a lousy song in order to improve it. You might be able to get something like Amazon's book-picker algorithm, though. i.e. if I like songs A1, A2, and A3, I'm likely to enjoy A4 but not B4. And maybe by evolving and abstracting rules for such software, we can find a way to make better musical composer AI.
Reply
Leave a comment