Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
From Prose to Rhyme: Automatic Synthesis of Rap Lyrics
This project explores automatic generation of song lyrics from different types of free form text, such as news articles or short stories.
Keywords: natural language processing, text generation, style transfer, song lyrics
Automatic rap lyrics generation [1, 2] is an interesting challenge for language generation systems, combining story development and creativity with verse generation that contains rhyme and flow. In this thesis, you will develop lyrics generation systems that use an existing text written in ordinary prose style (e.g. short stories or news articles) as a template for generation. To achieve this, we will rely on recent developments in **style transfer** of text (e.g. [3]), adopting them for the task where necessary.
**References**:
[1] Malmi, Eric, et al. "Dopelearning: A computational approach to rap lyrics generation." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016.
[2] Potash, Peter, Alexey Romanov, and Anna Rumshisky. "GhostWriter: Using an LSTM for automatic rap lyric generation." Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. 2015.
[2] Subramanian, Sandeep, et al. "Multiple-Attribute Text Style Transfer." arXiv preprint arXiv:1811.00552 (2018).
**Requirements**: good Python skills, knowledge in natural language processing
**Bonus points** if you enjoy listening to rap
**Starting date**: flexible
**Note:** we don't have any funding for Master theses. You will either need to self-fund, or apply for a scholarship independently.
Automatic rap lyrics generation [1, 2] is an interesting challenge for language generation systems, combining story development and creativity with verse generation that contains rhyme and flow. In this thesis, you will develop lyrics generation systems that use an existing text written in ordinary prose style (e.g. short stories or news articles) as a template for generation. To achieve this, we will rely on recent developments in **style transfer** of text (e.g. [3]), adopting them for the task where necessary.
**References**:
[1] Malmi, Eric, et al. "Dopelearning: A computational approach to rap lyrics generation." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016.
[2] Potash, Peter, Alexey Romanov, and Anna Rumshisky. "GhostWriter: Using an LSTM for automatic rap lyric generation." Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. 2015.
[2] Subramanian, Sandeep, et al. "Multiple-Attribute Text Style Transfer." arXiv preprint arXiv:1811.00552 (2018).
**Requirements**: good Python skills, knowledge in natural language processing
**Bonus points** if you enjoy listening to rap
**Starting date**: flexible
**Note:** we don't have any funding for Master theses. You will either need to self-fund, or apply for a scholarship independently.