With the help of this project, users will be able to choose a musical genre and then receive the lyrics of the song.
Model
I think we can use a pre-trained GPT2 model in English and fine-tune it to learn English lyrics.
The GPT2 model, pre-trained in English: link.
Datasets
We can parse the needed datasets (ex.: Genius API) or combine open-source datasets, and make them public
Training scripts
I known at least 2 public scripts to fine-tune a GPT2 model in Flax:
- From huggingface/transformers GitHub
- “Causal Language Model Training in TPU with Transformers & JAX”
Project outcome
The desired project output is a model that can generate a whole song by genre.
Challenges
- Create a cool dataset
- Allow choosing a genre
Optional
Also, it would be cool to allow users to choose an artist but there are some cons…
Conclusion
We can integrate this project with Video and Music generation so we can create our AI-Artist
I hope to find like-minded people and create a cool project