Skip to main content
Credits

Voice technology

Meta AI — Massively Multilingual Speech

Open speech-synthesis models

Kuku YalanjiAnindilyakwa

The open neural text-to-speech models the synthesized voices run on.

The MMS project (Vineel Pratap, Andros Tjandra and colleagues at Meta AI, "Scaling Speech Technology to 1,000+ Languages", 2023) released open speech-synthesis models for over 1,100 languages, trained on recorded readings of public-domain texts. The app uses their Pitjantjatjara model as a donor voice — read through a language-specific spelling bridge — for languages without a model of their own.