ZONOS3-Mini
Zyphra🇺🇸 USMajor lab· Nov 9, 2026
Audioopen weights
Compact open-weights text-to-speech model with about 100M active parameters. It clones voices from an audio sample, lets you design voices by age, accent and vocal texture, and runs in real time on phones, laptops and in the browser.
Summary of the model card
- License
- apache-2.0
- Modalities
- text → audio
2 variants
Sources
HuggingFaceDeutsch
Kompaktes Open-Weights-Sprachmodell mit rund 100 Millionen aktiven Parametern fuer Text-zu-Sprache. Es klont Stimmen aus einer Audioprobe und erlaubt Emotion, Tempo und Vortrag bis auf einzelne Woerter zu steuern, auch lokal im Browser.
Provenance
- Detected
- Nov 10, 2026
- First source
- huggingface
- Description
- Summary of the model card
- Editorially reviewed
- —