Dia 1.6B
VerifiedOpen-weights text-to-dialogue model that voices two speakers with laughs and pauses
Product preview
Dia 1.6B
What it is
Dia 1.6B is a 1.6 billion parameter text-to-speech model released by Nari Labs under the Apache 2.0 license. Rather than reading one line at a time, it generates a whole conversation in a single pass from a transcript that marks speakers with [S1] and [S2] tags. It can produce nonverbal sounds such as laughter, coughing and throat clearing, and it can condition on a short audio sample to copy a voice, tone or emotion. Weights are hosted on Hugging Face, the model works with Hugging Face Transformers, and a hosted demo Space is available. It currently supports English only and runs best on a GPU.
Key features
- Two-speaker dialogue generation in one pass
- Nonverbal sounds like laughs and coughs
- Voice cloning from a short audio prompt
- Apache 2.0 open weights on Hugging Face
What we liked
- + Free and open for commercial use under Apache 2.0
- + Natural back-and-forth conversational audio
- + Runs locally for full control
Worth noting
- · English only
- · Needs a GPU and some technical setup
Editor's notes
More context from AI BirdView
Dia 1.6B caught attention because a small team released an open model that handles conversation, not just narration. You write a script with speaker tags, and it returns audio where two voices trade lines with natural timing, plus laughs or sighs where you ask for them. For developers and podcasters who want to experiment without per-character fees, that is a rare combination.
- Generates multi-turn dialogue in a single pass from a tagged transcript
- Supports nonverbal cues such as laughter, coughing and throat clearing
- Clones a voice or tone from a 5 to 10 second audio sample plus its transcript
- Open weights under Apache 2.0, with Hugging Face Transformers support
- A free demo Space lets you hear results before installing anything
Dia 1.6B is best for developers, researchers and audio hobbyists who are comfortable with Python and want a local dialogue voice model. A practical tip: follow the project's guidance on input length, since very short scripts can sound unnatural and very long ones speed up, and alternate speaker tags strictly for the cleanest output. Nari Labs has since released a successor, Dia2, which is worth a look too.
Reader reviews
From people who've used Dia 1.6B. Emails are never shown publicly.
No reviews yet. Be the first to share your experience.
Alternatives to Dia 1.6B
Other tools we like in this space.
ElevenLabs
The most lifelike AI voices on the market
Resemble AI
Deepfake detection and AI watermarking, plus free open-source Chatterbox voice models
Cartesia
Real-time voice AI with 40ms latency
WellSaid
Enterprise voiceover with voices built consentfully