What it does Feed it any artist’s lyrics and it spits out new bars, word by word, that rhyme and sort-of flow. Under the hood, a Markov chain generates candidate lines, then a recurrent neural net picks sequences by desired rhyme and syllable count. There’s also a speech.py mode that performs the results over a generic beat using text-to-speech — the author admits it sounds “pretty trash.”

The interesting bit The whole thing was a high school programming club project by a West Virginia teen who taught himself neural nets. The README’s future goals section is endearingly ambitious: hiring a Kanye impersonator to train a voice clone, then “an AI resurrection of dead rappers” like Biggie and Big L.

Key highlights

• Ships with Kanye’s full discography pre-loaded in Lyrics.txt
• Swap in any artist by replacing the lyrics file and toggling train_mode
• Outputs to terminal and neural_rap.txt; speech.py adds robo-performance
• Google Colab notebook available for zero-install runs
• documented_model.py includes inline comments explaining the model

Caveats

• Requires Python 2.x, which is end-of-life
• Generated lyrics include vulgarities and occasional nonsense (“Mayonnaise colored Benz”)
• No candidate images available in the repository

Verdict Worth a spin if you’re teaching generative text or want a concrete, small-scale RNN example. Skip it if you need production-grade lyric generation — this is art-project code, not a framework.