Skip to content

MaskGCT

Zero-shot TTS without explicit alignment

Visit official site

What MaskGCT is

MaskGCT is described as a zero-shot text-to-speech model built on a masked-and-predict learning paradigm with two stages.

What you can do with it

  • Predict semantic tokens from text
  • Predict acoustic tokens from semantic ones
  • Imitate a voice style and emotion from a prompt

Application scenarios

  • A researcher makes a character voice sample
  • A firm makes multilingual service replies
  • A creator makes audiobook narration

Pros & cons

✓ What we like

  • Prosody, style and emotion follow a prompt voice
  • Total audio duration can be controlled

! What to watch out for

  • Users adjust emotion, style and pace parameters
  • The material reports quality and similarity results

FAQ

How many stages are there?

The material describes a two-stage model.

Is extra training needed?

The material states zero-shot in-context learning without extra training.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review