Audiobox
Meta FAIR's research demo for unified speech, voice cloning, and sound generation.
About
Audiobox is a research model from Meta's FAIR lab that generates speech, sound effects, and ambient soundscapes from text descriptions in one unified architecture. You can type a sentence for a cloned voice to speak, describe a voice ("deep feminine voice") and have it synthesized from scratch, restyle a recording of your own voice, or prompt sounds like dogs barking or children playing. It is built on flow-matching with a self-supervised pretraining objective, and it set new benchmarks on speech similarity and text-to-sound quality at release.
It is strictly a research demo. Meta does not allow commercial use, it is not open source, and the demo was historically restricted in Illinois and Texas over biometric privacy law. Availability depends on Meta keeping the demo servers up, so treat it as a look at the research rather than a tool you can build on.
What it can do
- Text-to-speech with novel voice styles
- Voice cloning from short recordings
- Text-to-sound effect generation
- Voice restyling and characteristic guidance
- Unified speech and sound model
Setup
Open Meta's Audiobox demo page and use it in the browser. No install needed.
Prerequisites
None beyond a browser; not available for commercial use.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 52; Trending score 50/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.