Skip to content

Dia AI

Generate dialogue with speaker tags and nonverbal sounds

Visit official site

What Dia AI is

Dia is described as a text-to-speech model from Nari Labs with 160 million parameters that produces dialogue directly from text. Pretrained model weights are hosted on Hugging Face, and the listed generation target is English.

What you can do with it

  • Separate speakers with [S1] and [S2] tags
  • Generate nonverbal sounds such as laughter and coughs
  • Clone a voice by uploading audio
  • Control emotion and tone through audio conditioning

Application scenarios

  • Generate dialogue content for a virtual assistant
  • Create varied voices for game characters
  • Produce spoken narration for educational videos

Getting started

  • Clone the repository from GitHub
  • Enter the dia directory
  • Install dependencies with pip install -e .
  • Launch the Gradio UI with python app.py

Pros & cons

✓ What we like

  • Audio conditioning controls emotion and tone
  • A Gradio UI is provided for interaction

! What to watch out for

  • The listed generation target is English only
  • Setup is documented through cloning the GitHub repository and installing dependencies

FAQ

How are speakers distinguished?

The material names [S1] and [S2] tags for separating speakers.

Where are the pretrained weights?

The material states the pretrained model weights are hosted on Hugging Face.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review