subpath / keras_music_gereration

Music generation with Keras and LSTM

Keras Music Generation

Music generation with Keras and LSTM

Requirements

Python

1. Jupyter Notebook
2. Python3.7

Packages

mido==1.2.9
tensorflow==1.14.0
Keras==2.1.2
sklearn==0.0
numpy==1.17.3

Check Jupyter Notebook

keras_music_gereration's Issues

the output midi gets slower and slower

some error at the end of the code

ValueError: Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 1 array(s), but instead got the following list of 8734 arrays: [array([[-0.61290323, 0.42222222]

it is laocation is about line 101 i have no idea how to fix it

#line 99 mid = MidiFile()
#line100 track = MidiTrack()
#line 101
#line 102 t = 0

Provide more information on each chunk of code in Notebook

Hi dude,

First of all thanks a lot for your work. Now I wanted to ask you if it would be possible for you to provide more information on each chunk of code in the notebook of your repo ?

Thanks !

Regards,

Loss is NaN

While training, the loss is NaN all the time. Logging the code in the earlier steps, I figured that the combine variable you are using to combine notes and velocities has a NaN in every element.

It becomes NaN after this block:

note_min = np.min(notes)
note_max = np.max(notes)
velocities_min = np.min(velocities)
velocities_max = np.max(velocities)

for i in combine:
    i[0] = 2*(i[0]-((note_min+note_max)/2))/(note_max-note_min)
    i[1] = 2*(i[1]-((velocities_min+velocities_max)/2))/(velocities_max-velocities_min)

combine new value:

[[-0.3333333333333333, nan],
 [0.3333333333333333, nan],
 [-0.3888888888888889, nan],
 [0.2777777777777778, nan],
 [-0.5, nan],
 [0.16666666666666666, nan],
 [-0.6111111111111112, nan],
 ...,
 ...,
 ...,
]

Caused by division by 0. (velocities_min is equal to velocities_max, so their difference becomes 0.)

generated midis empty?

I downloaded the midis generated here, Generated_song_epoch=10.mid but they were empty