Bars.ai

Training a neural network to generate song-lyrics using a number of curated datasets containing lyrics from music artists.

by Alex Pinette, Milo Lani-Caputo, and Tamsin Rogers

View our blog post about this project HERE

Scraping data

Using Genius, and Spotify Web APIs, we are able to get song lyrics as well as playlist and artist data.

Getting started

First:

git clone https://github.com/alexpinette/Bars.ai.git

Then you are going to need to install the required libraries:

pip install -r requirements.txt

Create your own API client

Genius

• Follow the instructions on Genius API Documentation to get your own client_id, client_secret, and client_access_token.

• Then create a genius_config.py file with variables for your client_id, client_secret, and client_access_token.

Spotify

• Follow the instructions on Spotify API Documentation to get your own client_id, client_secret.

• Then create a spotify_config.py file with the variables for your SPOTIPY_CLIENT_ID, and SPOTIPY_CLIENT_SECRET.

How to use

Data Generation

• Open the data_generation_spotify or data_generation_genius notebook.

• Enter a playlist/album URL in the allocated variable for data_generation_spotify.

• Or, enter an artist name in the allocated variable for data_generation_genius.

• Other options are to choose from the already created datasets.

Training the Model

• Open the model_generation_and_training notebook.

• Input the currated .csv file in the associated variable.

• Run the notebook.

Generating New Lyrics

• Open the lyric_generation notebook.

• Update the associated variables with the correct dataset, and model.

• Create your own or allow the script to randomly generate a seed.

• Run the notebook.