ZONOS3-Mini / README.md
Gatozu35's picture
Deploy ZONOS3 a50e141-dirty
fa42cb8 verified
|
Raw History Blame Contribute Delete
1.27 kB

A newer version of the Gradio SDK is available: 6.32.0

Upgrade
metadata
title: ZONOS3 Mini
sdk: gradio
sdk_version: 6.17.3
python_version: 3.12.12
app_file: app.py
pinned: false
short_description: Multilingual TTS with voice cloning and steering controls
startup_duration_timeout: 1h
models:
  - Zyphra/ZONOS3-Mini

ZONOS3 Mini

The Zonos Voice Studio (Cast, Direct, Edit and the lite page) over the plain-PyTorch reference implementation of ZONOS3 Mini, with every take and every speech edit generated on ZeroGPU. Pronunciations can be typed in X-SAMPA between double square brackets, "I met [[S@vQn]]" (the release phoneme tokens), and pauses as <0.5> after a word. "Phonemize selection" looks words up in the gold pronunciation lexicon (lexicon/); words it lacks are guessed with espeak-ng (GPL-3.0, through espeakng-loader). Projects stay in your browser. LACI (Project settings, or Advanced on the lite page) regenerates a take from where it skipped words, babbled or stopped early.

The lexicon is built from WikiPron (Wiktionary, CC BY-SA), CMUdict, ipa-dict and Britfone; lexicon/manifest.json lists every source file with its URL and SHA-256.

The first start downloads about 10.5 GB of weights, so it takes several minutes.

Source: servers/space/ in Zyphra/ZONOS3, deployed with servers/space/deploy.py.