#BringbackFranku
This project aims to produce an AI that can generate new Filthy Frank scripts with a stretch goal of having those scripts synthesized with Franku's voice.
Goals
- Evaluate the best NLP framework for the job
- Test this initially on movie/music scripts.
- Collate all the Filthy Frank episode dialogue into text files.
- Generate new scripts.
- Work on voice synthesis
Low Level
Once the initial scripts have been colalted from using youtube -> captions -> edit and saving to file: sentences will need to be tokenised. Most NLP processes work best on smaller chunks rather than episodes at a time. The challenge will setting delimiters, token values, and associated meta-data.
Potential MetaData design
| Episode | Length | special words | characters | languages | text |
|---|---|---|---|---|---|
| PINK GUY COOKS FRIED RICE AND RAPS | 14 | slurp noises | pink guy | English, Japanese | I ain't no Gordon Ramsey but I make the best damn rice ya pansies. |
Tools
A list of suggested tools to work with.
- word2vec
- nltk
- Spacy for preprocessing and tokenisaiton
- LSTM keras
Links that have beeen helpfull
https://www.kaggle.com/jrobischon/wikipedia-movie-plots/kernels
https://juanitorduz.github.io/movie_plot_text_gen/