🗣️ Dia: New Open Source Text-to-Speech Model

Nari Labs, a two-person startup, has launched Dia, an open-source text-to-speech model. This model, boasting 1.6 billion parameters, is designed to generate natural-sounding dialogue from text, even incorporating emotional tones and nonverbal cues. Its creators claim Dia surpasses existing proprietary models from companies like ElevenLabs and Google in terms of quality and nuanced control. The model's code and weights are freely available, allowing developers to download and deploy it locally. Dia supports features such as speaker tagging and the interpretation of nonverbal cues within the text prompts, offering more customizable speech generation. Nari Labs provides comparison examples highlighting Dia's superior performance in dialogue scenarios, emotional delivery, handling of nonverbal cues, and even rhythmic content like rap lyrics. Distributed under an Apache 2.0 license, Dia is intended for various applications, from content creation to assistive technologies, with a focus on ethical use and community collaboration.