Docker Desktop or Docker Engine. An NVIDIA GPU is optional; CPU mode works too. The full image includes all five model packs.
Fast local speech Stable v1.0 or rolling latest
KokoroTTS
Natural speech, running on your own computer.
A local-first 82M-parameter TTS workspace with 173 voices across 11 languages, five independently managed model packs, streaming, SSML dialogue, OpenAI-compatible APIs, and opt-in MCP tools.
Open Generate, type a sentence, choose a voice, and select Generate. Use Stream for cancellable synthesis, API for integration status, and System for model-pack and GPU controls.
The full image can run without a model download after it has been pulled. No text or generated audio needs to leave your computer.
What is included
A complete local workspace, not just a model.
- 173 voices across 11 languages
- Five independently managed model packs
- Generate, Stream, API, and System views
- Seven audio formats and cancellable streaming
- OpenAI-compatible and native APIs
- SSML dialogue and opt-in MCP
Voice and model catalog
One workspace, five model families.
The v1.0 release and rolling latest combine standard Kokoro voices with dedicated German, Vietnamese, and enhanced Chinese checkpoints. Model packs can be enabled independently and load only when used.
54 standard voices, 2 dedicated German voices, 14 Vietnamese voices, and 103 voices from the Chinese v1.1 pack.
Eight standard Mandarin voices plus 100 enhanced Mandarin speakers.
Standard Kokoro, German Martin, German Victoria, Vietnamese, and Chinese v1.1.
American and British English, Japanese, Mandarin, Spanish, French, Hindi, Italian, Brazilian Portuguese, German, and Vietnamese.
Kokoro v1.1 Chinese
A dedicated model for Chinese speech.
The pinned Kokoro v1.1 Chinese checkpoint expands Mandarin choice substantially while preserving the original standard voices. It is managed as its own model pack inside the same KokoroTTS workspace.
Hear the Chinese voices ↗- 100 enhanced Mandarin voices: 55 female and 45 male.
- 108 Mandarin choices in total when combined with the eight standard voices.
- Three additional English voices support mixed Chinese-English synthesis.
- The full image includes the model for offline use; the tiny image downloads it when first selected.