CHOOSEConsumer & creative
Ludwig van Jev-thoven!
Srivatsa Rampalli@svatsa159
0:28This project uses a classifier called 'jev' to generate musical phrases note by note by selecting from probabilistic options, rather than generating text.
Signal@typesafeai's jev doesn't generate text, it just picks from options with probabilities.
Ludwig van Jev-thoven!
Gave a classifier a piano. @typesafeai's jev doesn't generate text, it just picks from options with probabilities. So I asked it, note by note: which key, how long, which chord, staccato or tenuto, is the phrase over? 20 to 30 calls, ~370 ms each, and out comes an 8-bar wistful Andante in C: a I–ii–IV–vi–ii–IV–V7–I progression, grace notes, ties, phrase slurs, and a whole-note C under a fermata to finish. Sampling its distribution instead of its top pick means every run is a different tune. Rendered live with VexFlow.