This repository contains the source code and dataset used in our study on the application of sequence-to-sequence deep learning models for generating rap lyrics that stylistically reflect the music of Drake. We have explored models including Recurrent Neural Networks (RNNs), XLNet, and GPT-2, focusing on their ability to handle the complexity of linguistic patterns and thematic elements found in Drake's lyrics.
The aim of this project is to bridge the gap between artificial intelligence and artistic creativity, enabling the generation of rap lyrics that not only mimic Drake's style but also push the boundaries of automated creative writing. Our study evaluates these models based on qualitative assessments and quantitative metrics such as BLEU, Kappa score, and Cosine Similarity.
- RNN: Basic sequential processing for initial baseline lyrics generation.
- XLNet: Advanced model with permutation-based training designed to handle bidirectional contexts effectively.
- GPT-2: Transformer-based model with superior performance in handling complex dependencies and stylistic nuances.
For any further queries or discussions, feel free to contact us:
- Vatsal Bagri (vatsal.bagri@mail.utoronto.ca)
- Maxim Radulovic (max.radulovic@mail.utoronto.ca)
- Rayan Hossain (rayan.hossain@mail.utoronto.ca)