BDB Bertsolaritzaren datu-basea

We present a human-in-the-loop GAN framework for battle rap, where a human artist (MC) serves as generator, and the AI acts as an adaptive discriminator. The AI provides feedback on rhyme complexity, coherence, and stylistic alignment, challenging the MC’s improvisational skill. Fine-tuned language models emulate diverse rap styles, while voice cloning creates adversarial loops: the MC competes against stylised versions of their own voice in a dynamic, selfreflective duel. The system follows a dual-phase process: (i) an Emulation Phase, where AI mimics established flows to reinforce technical mastery, and (ii) an Improvisation Phase, where AI disrupts expectations to prompt originality. This ensures that creative growth emerges from constraint and challenge. Success is judged through MC evaluations of the AI’s performance as an adversary. Framed as a study paper, this work offers a thought experiment in adversarial co-creativity, modelling how AI might inspire, rather than merely assist, human expression. Beyond computational modelling, the framework offers insights into machine-mediated creativity and how AI can be designed to provoke human creativity through improvisation, challenge, and real-time performance. The study positions the AI as a dynamic co-performer capable of eliciting novel artistic responses. As such, it contributes to emerging discourse on creative AI systems that influence, not just assist, human expression.