k2-fsa/ZipVoice

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1045 stars · 143 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

ZipVoice is a series of fast, high-quality zero-shot text-to-speech models based on flow matching, with a compact 123M-parameter Zipformer-based architecture. It supports voice cloning in Chinese and English, single-speaker and dialogue generation, and includes a distilled variant for faster inference.

Use cases

• clone a voice from a short audio sample
• generate speech from text in Chinese or English
• generate multi-speaker dialogue audio
• run fast TTS on CPU or GPU with few sampling steps
• build an audiobook or narration pipeline with cloned voices
• generate cross-lingual speech with a speaker prompt

When to choose

• you need fast, small-footprint zero-shot TTS with high speaker similarity
• you need Chinese and English speech synthesis or dialogue generation
• you want an open-source Apache-2.0 TTS model with checkpoints and inference code

When to avoid

• you need languages beyond Chinese and English
• you need a managed cloud TTS API rather than self-hosted models
• you need real-time streaming synthesis out of the box

Facets

library · maturity active

tts machine-learning deep-learning speech-recognition speech-processing machine-learning python cross-platform zero-shot-tts voice-cloning flow-matching zipformer speech-synthesis multilingual natural-language-processing gpu

Member repositories

For agents

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