MongoDB - Build AI That Scales
Mini toolsDia 2
By Nari Labs
Dia2 is built for conversational text to speech where you do not have the full sentence in advance. It can start speaking as soon as the first words arrive, continue generating up to about two minutes of English audio, and optionally condition on previous audio turns so an assistant can keep a consistent voice and react smoothly to the user. The project provides 1B and 2B models plus simple CLI and Python APIs, and recommends prefix conditioning or fine tuning if you want a stable, specific voice, since the base model intentionally varies timbre between generations. Licensed under Apache 2.0, Dia2 is aimed at research and real time dialogue systems that need open weights and low latency speech generation.
New
Audio
Gen 4
Overview
Dia2 is an open source streaming dialogue TTS model that generates speech in real time from partial text, supports audio conditioning for natural back and forth conversations, and ships 1B and 2B checkpoints under Apache 2.0.
About Nari Labs
Open-source TTS startup behind Dia, a 1.6B real-time, zero-shot voice-cloning model.
Website:
narilabs.org
Tools using Dia 2
No tools found for this model yet.
Last updated: February 25, 2026
