Translating lyrics to chords with RNNs
- 1 minThis is a project carried out for the Computer and Music course at EPFL by Pr. Süsstrunk.
Made in collaboration with Matheus Bernat, and Husam Jubran.
We human beings have a great power of creativity, and machines have a great power of extracting hidden patterns and replicating them, so why not join these two powers? In this project we aim to help uninspired songwriters who have the perfect lyrics but want a chord progression that goes well with the lyrics. To do this, we constructed a dataset of English songs with nearly 30 thousand songs from Ultimate Guitar, perform chord and lyrics analyses on it, and then construct a recurrent neural network (RNN) that takes in lyrics and outputs a 4-long chord progression. As the main results of this project, we have the dataset whose features can be further explored, and a Jupyter notebook with the RNN that takes in lyrics and outputs the 4-long chord progression. —
About this project
- Python, Javascript
- Packages/Librairies: Pandas, skLearn, plotly, TensorFlow, ReactJS, Flask
- Interactive webapp, identity and topic oriented analysis of the QuoteBank dataset
Links
- Link to the report.