ASSIGNMENT 2 (due May 10 or 17).

Generate a random-walk solo over a 12-bar blues chord progression. If you don't like the 12-bar blues, make up your own chord progression or perhaps generate one with a Markov chain. This is an open-ended assignment, and it will be quite hard to make really excellent blues, so be prepared to stop before you get quite the result you imagined.

A possible way to proceed would be to make a simple random walk and then quantize it down to a scale that changes with the chord. Problem: random walks tend not to land on good spots at downbeats. There are various ways you could try to affect things to work better than a pure random walk.

Alternatively, make or find a corpus of reasonable solo material you can stitch together to make the solo. This could be done recombinantly (see Gerard Assayag's talk May 26), or probabilistically (Markov chain with various possible tweaks) or using a grammar (Bob is more of an expert on that possibility than I am).

Alternatively, train a neural net to "predict" each note given the harmony and the 2 or 3 (or more) previous notes.

Alternatively, set it up as a rule-based/constraint or optimization problem. (Unfortunately we won't be covering these before May 10 so this might be hard to do.)

Here is my own attempt ( Wav, 2.5M ). The bass and the 'solo' are both random walks, each with its own rhythm and scales.