This is a work in progress and I don't currently have any good results.
Experiments in generative music using high order layers. Generated audio and waveforms can be seen in the tensorboard output.
Free music archive tool is located here and the
Make sure to install ffmpeg as it's used by torchaudio. When running things on the command line
poetry install
and then
poetry shell
Working on memorizing this musics Tea K Pea - vpunk
Try and memorize as set using conv net
python examples/memorize.py
with a tail focus network
python examples/memorize.py net=tail_focus
to try with shorter versions
python examples/memorize.py data.max_size=100000
and for debugging
python examples/memorize.py data.max_size=10000 data.window_size=1000
and
python examples/memorize.py net=tail_focus_small data.max_size=10000 data.window_size=1000
write out an associated mp3
python examples/generate.py checkpoint=\"outputs/2022-07-14/07-14-52/lightning_logs/version_0/checkpoints/epoch=49-step=1800.ckpt\"