Skip to content

EMOVA

An emotionally rich multimodal voice assistant model

Visit official site

What EMOVA is

The model is a multimodal language model that processes speech end to end while keeping vision-language ability. A decoupled speech tokenizer enables dialogue with rich emotion.

What you can do with it

  • Handle vision and speech inputs together
  • Control emotion and pitch of the voice
  • Converse with vivid emotional expression

Model notes

  • Vision-language results are described in the material
  • ASR performance is described as state of the art
  • No external tools are needed for image, text and speech

Pros & cons

✓ What we like

  • Emotion and pitch can be controlled
  • Vision-language and ASR benchmarks are reported

! What to watch out for

  • Access starts from the official site
  • The material focuses on a research demo

FAQ

What does the name stand for?

EMotionally Omni-present Voice Assistant is given in the material.

What is reported on benchmarks?

The material states it surpasses GPT-4V and Gemini Pro 1.5 and approaches GPT-4o.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review