k2-fsa/OmniVoice

Health v2 · maintenance only

Adoption not part of the score

14762 stars · 2203 forks observed · 2026-10-19

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

OmniVoice is a massively multilingual zero-shot text-to-speech model supporting 600+ languages, built on a diffusion language model-style architecture. It provides high-quality voice cloning, voice design via speaker attributes, and fast inference (up to 40x real-time) through Python APIs and command-line tools.

Use cases

• clone a voice from a short audio sample
• generate speech in hundreds of languages from text
• design a custom voice by specifying gender, age, pitch, and accent
• synthesize speech with fine-grained control like [laughter] tags and phoneme pronunciation correction
• run fast TTS inference on GPU or Apple Silicon
• fine-tune or evaluate a multilingual TTS model

When to choose

• you need zero-shot TTS with the broadest language coverage (600+ languages)
• you want state-of-the-art voice cloning quality
• you need very fast inference (RTF as low as 0.025)
• you want fine-grained control over non-verbal sounds and pronunciation
• you prefer an Apache-2.0 licensed model with Python API and CLI

When to avoid

• you need a lightweight CPU-only TTS without GPU acceleration
• you need a simple single-language TTS with minimal setup
• you want a hosted SaaS API rather than running models locally
• you need real-time streaming synthesis specifically

Facets

library · maturity active

tts speech-recognition machine-learning deep-learning audio-processing speech-processing machine-learning artificial-intelligence python cross-platform cli voice-cloning zero-shot-tts multilingual diffusion-language-model voice-design speech-synthesis natural-language-processing gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="k2-fsa/OmniVoice")