Skip to content

Stability AI text-to-speech models

Natural language guided high fidelity text to speech

Visit official site

What Stability AI TTS is

The models are described as high fidelity text to speech models that use natural language guidance, with labels over speaker identity, style and recording conditions.

What you can do with it

  • Guide speech synthesis with natural language
  • Label speaker identity, style and recording conditions
  • Train a speech language model on a 45000 hour dataset

Example uses

  • A female voice with a US accent for announcements
  • A male voice with a UK accent for recording
  • A male voice with a South African accent for announcements

Pros & cons

✓ What we like

  • Simple methods for higher audio fidelity are proposed
  • Accent and speaker choices appear in the example uses

! What to watch out for

  • Guidance depends on labels over the training data
  • The described output relies on the discovered dataset

FAQ

What is labelled?

Speaker identity, style and recording conditions.

How large is the dataset?

The material states 45000 hours for training.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review