Lyrix

Scenario 1: Train a CBOW model on song lyrics of one genre, then remove random words from these lyrics, and predict the closest word from another genre (e.g., replace all the cursing words from rap songs with respective words from kid’s songs). In order to achieve this, you will have to train different word embeddings by genre, and then align them (e.g., using MUSE).

Scenario 2: Style transfer from one genre to another (you can check the related work on images).

The project requires programming skills in Python. Relevant experience in Natural Language Processing and Deep Learning is preferred.