This is a Rap generator that has adjustable styles, lengths, and "temperature" aka randomness. Optionally, the user can train new models that will match the style of any input text (this takes some time). Also, the user can input a starting text, similar to talk to transformer
We have access to lyrics from all publicly available songs.
Access the Genius.com API through the lyricsgenius python package to get lyrics from prolific rappers, then train models to learn the unique styles of each.
Take lyrics from all songs by each rapper and place them in unique .txt files to be used when fine tuning the model. Put lyricsgenius functions into .py files for simplicity.
We used google colab with a gpu to Retrain gpt-2's 345mb model on each of the rappers, then saved each uniquely retrained model for future use.
We have a simple generator that can make relatively cohesive raps based off of the chosen rapper's style and the prefix text.
Ideally deploying the auto rapper to the web as a flask app would be nice, also increasing the number of rapper models
We borrow heavily from the code in this repo and this google colab which is also linked in the repo.