Two undergrads built an AI speech model to rival NotebookLM

Two undergraduate students have developed an open-source AI model called Dia, inspired by Google’s NotebookLM, which can generate podcast-style clips with customizable voices and scripts. The model, built using Google’s TPU Research Cloud for free training, offers features like voice cloning and non-verbal cues, running on standard PCs. While it performs well in generating conversations, concerns about potential misuse and the use of copyrighted content for training have been raised by critics. Nari Labs aims to expand Dia into a broader platform with future models and additional functionalities.