plunkgj / midi2wave Goto Github PK

I found your project and it seems related to my idea that I posted on a different repo here facebookresearch/music-translation#5
I will appreciate if you take a look at the idea and see if it is possible to implement.
It seems that in this project you tried to achieve similar results as I described there, but maybe you didn't think of converting monophonic midi to monophonic instruments like sax/clairnet/trumpet/violin. Trumpet is much more similar to human voice than piano, maybe you could revisit your project with this different assumption and it might be successful.

Hi, I hope you are doing good.
first of all thank you for that great job it's very helpful,

I am trying to reproduce the results for educational purposes and I 've faced some problems , tried to solve them but now i 'm stuck in here so i thought i can ask for some help .
due to limited resources i am using VGMidi dataset instead of the maestro dataset (if this itself will cause a problem or so please tell me ) i created the csv file and did the first step resampling and preprocessing
now i am in the train step and i get this error each time i try to train

the number in entry 1 changes from run to run , 120000 is length of what so maybe i can think of what causes the length to change.
another important thing to notice is that the config_train.json had2 missing attrivutes use_skip_out and same_cond_each_resblock so i added them and assumed them both to be true , also the order of the "use_cond_act": false,
"use_cond_bias": true, attributes is reversed in the function train so is it ok or should we reverse them