Skip to content

LLaMA-Omni

Low-latency speech interaction with text and audio

Visit official site

What LLaMA-Omni is

LLaMA-Omni is described as a low-latency, end-to-end speech interaction model built on Llama-3.1-8B-Instruct. It targets GPT-4o level voice ability and produces text and speech responses at once.

What you can do with it

  • Hold a low-latency voice conversation
  • Generate text and speech together
  • Try the model through a Gradio demo
  • Run local inference scripts

What to watch out for

  • Training used 4 GPUs in under 3 days
  • Latency is stated as low as 226 ms
  • Two model downloads are named

Pros & cons

✓ What we like

  • A Gradio demo is provided
  • Local inference scripts are included

! What to watch out for

  • Setup names extra libraries
  • Two model downloads are listed

FAQ

How low is the latency?

The material states latency as low as 226 ms.

What is it built on?

Llama-3.1-8B-Instruct is named as the base.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review