Sounds interesting; harmonically it is correct, but temporally there is room for improvement (e.g. adherence to measure lengths and form over the entire number of measures); apparently MIDI data is used for training and is generated (correct me otherwise) so it is comparable to https://openai.com/blog/musenet/; is there a publication somewhere explaining the details?
1 comment
[ 2.9 ms ] story [ 14.4 ms ] thread