Ian Simon
Ian Simon
As a quick temporary solution, change line 40 of setup.py to `'librosa >= 0.6.2, < 0.7',`. Not sure what to do going forward as I also was unable to get...
We should have a colab out in the next couple days. Sorry about the delay!
I'm looking into the issue. For now you can try adding a line `pip install --upgrade jaxlib` to the first cell.
Oops, that will remove CUDA support. The line should be: `!pip install --upgrade jaxlib -f https://storage.googleapis.com/jax-releases/jax_releases.html`
Downloads in colab can be a little flaky. Does it work if you try the same download again?
Unfortunately there is currently no PerformanceRNN data converter for use with MusicRNN. You could try adapting the (totally separate) Performance RNN JS demo codebase to work with your checkpoint: https://github.com/magenta/magenta-demos/tree/master/performance_rnn
The issue seems to be that the notes have no velocity. If I manually set the velocities the conversion works.
Depending on what the MIDI track looks like, you can try one of the following two functions: https://github.com/magenta/note-seq/blob/main/note_seq/sequences_lib.py#L1668 (if the MIDI track consists of block chords) https://github.com/magenta/note-seq/blob/main/note_seq/chord_inference.py#L260 (if the MIDI...
Should be fixed now.
Ugh, I'm not sure the best way to fix this. For now, you can add another pip install command that looks like: ``` !python3 -m pip install clu==0.0.8 numpy==1.22.4 --ignore-installed...