dia

Why we included this project

Dia is an open-weight text-to-speech model from Nari Labs that treats dialogue as a first-class output rather than a side effect of reading text aloud. Give it a transcript of a conversation and it generates the entire exchange as one continuous output, complete with emotion, tone, and nonverbal sounds like laughter, coughs, or a cleared throat. That makes it a natural fit for voice acting, multi-character audiobook scenes, game dialogue, and any narration where expressive delivery matters more than a neutral announcer voice. You can also condition the output on a short audio clip to steer the emotion and tone of a line. The 1.6B weights are Apache-2.0 licensed and run through Hugging Face Transformers, with a hosted demo Space so you can hear what it does before wiring it into your own pipeline.