jsrp

The ImprovBot is a program that replaces artistic decisions with random selections of notes based on a previously saved scales. Its aim is to create music but it has to be flexible enough to work with other musicians, or songs (focusing on factors such as key, tempo, rhythm). Although in its infancy at the moment the program will aim to be able to make decisions based on a number of other elements. Programs such as allow users to keep tweeting after death by measuring the rate and frequency at which they ‘tweet’ things, this program could do the same but with music. The program will look at imitating musicians, creating a formula that will measure common interval, time and mood changes in songs. Then using this information along with random musical changes, common changes (this will depend era of the music) and influences on the musician it is trying to imitate.

Basically creating a sort of intuition for a Max patch.

The ImprovBot is a program that replaces artistic decisions with random selections of notes based on a previously saved scales. Its aim is to create music but it has to be flexible enough to work with other musicians, or songs (focusing on factors such as key, tempo, rhythm). Although in its infancy at the moment the program will aim to be able to make decisions based on a number of other elements. Programs such as  allow users to keep tweeting after death by measuring the rate and frequency at which they ‘tweet’ things, this program could do the same but with music. The program will look at imitating musicians, creating a formula that will measure common interval, time and mood changes in songs. Then using this information along with random musical changes, common changes (this will depend era of the music) and influences on the musician it is trying to imitate.

Basically creating a sort of intuition for a Max patch.


improvbot

Nice pretty funky patch and dig the route of investigation, may I suggest looking at the prob object, which is a weighted transition table. There's some pretty clever magic that goes on inside there, that will aid in decision making. I would suggest looking into the subject of machine learning, genetic algorithms, neural nets and markov tables for inspiration on where to take your patch, each of those have tried and tested and can make informed choices given certain stimuli. Or perhaps fuzzy logic could assist you equally.

The problem with random is that it's random, you have no control over the choices it makes, with [prob] object you can set weights to choices so that they will appear more frequently, so instead of everything being selected with equal probability you weight certain choices, for instance in your patch for making melody, the most frequent choices could likely be the 3rd,5th,7th's so you would weight them higher than the 9th, 13th etc, it's like fixing a dice, putting a weight on one of the faces so it will more than likely fall on the one you weighted down. So you could also change the weights variably based on certain contexts, I would suggest start mapping out decision trees just on paper or a white board, and explore how to emulate rational decisions you make in approaching improvising over chords using a set of weighted probabilities. This is the basis of neural nets, you don't have to make a neural net in the purest sense, but potentially taking a little inspiration from they way our brain makes choices based on stimulus/context/bias could be the way of solving this problem.

Equally using markov chains, there are a few decent implementations knocking around the net, very fun to look into as well!

Nice pretty funky patch and dig the route of investigation, may I suggest looking at the prob object, which is a weighted transition table. There's some pretty clever magic that goes on inside there, that will aid in decision making. I would suggest looking into the subject of machine learning, genetic algorithms, neural nets and markov tables for inspiration on where to take your patch, each of those have tried and tested and can make informed choices given certain stimuli. Or perhaps fuzzy logic could assist you equally. 

Equally using markov chains, there are a few decent implementations knocking around the net, very fun to look into as well!
