An NLP application that learns to dream up lyrics in the style of various indie artists or groups.
You can either provide the start of a sentence for the model to complete, or play a game of "recognise the lyrical style".
Setting up a server and storage for the data and models costs money so I stopped. You can still run the application locally, here's how:
- I stored the checkpoints using Git LFS, so you need to install it first. For instance on Debian and Ubuntu, call
sudo apt-get install git-lfs. See the complete instructions here. - Clone the repo. This will take a few minutes due to the large files.
python3 -m venv .venvsource .venv/bin/activate(or.\.venv\Scripts\activateon Windows)- Install PyTorch according to the official instructions here. The code was developed and tested on PyTorch 2.0.0; if you have problems with a more recent version, try downgrading to that one (see here). You don't need the GPU version to run the application.
pip install -r requirements
- Execute
flask run - Enter the following URL in a browser:
http://127.0.0.1:5000 - Have fun!
The underlying models are GPT-2s fine-tuned on each artist's collection of lyrics, which I found on Aleksey Korshuk's repo here.
The generated lyrics don't always make sense and sometimes overfit to existing lyrics, but you can definitely tell the difference between Bob Dylan and Florence+The Machine!