News Summary

Two undergraduates from Korea, co-founders of Nari Labs, have developed an AI speech model named Dia, which rivals Google’s NotebookLM. Despite their limited AI expertise, they leveraged Google’s TPU Research Cloud to train Dia, which boasts 1.6 billion parameters allowing it to generate podcast-style clips with customizable voice features. Available on platforms like Hugging Face and GitHub, Dia is capable of voice cloning and generating dialogues with various nonverbal cues. However, like many voice generators, it raises concerns over potential misuse for disinformation or scams, although Nari Labs discourages such actions. The founders plan to expand its language support and develop a social platform around Dia.