Zonos

Zonos is a local text-to-speech model for people who want to generate speech in a reference speaker's voice. It suits projects that need control over vocal delivery as well as voice matching.

• Voice cloning: Generate speech from text using a short recording of the speaker you want to match.
• Delivery and emotion: Adjust speaking rate, pitch variation and audio quality, and guide emotions such as happiness, sadness, fear and anger.
• Audio continuation: Supply an audio prefix to guide speaker matching and delivery, including whispering that a voice sample alone may struggle to reproduce.
• Multilingual speech: Generate speech in English, Japanese, Chinese, French and German.
• Browser interface: Use the included Gradio interface to generate speech through a web UI on your own machine or server.

Zonos is open source under Apache 2.0 and runs on Linux and macOS, with Docker deployment supported. Local generation uses your hardware; Zyphra also offers a separate hosted audio playground.

The model has transformer and hybrid variants. GPU execution requires at least 6 GB of VRAM, and the hybrid variant also requires an NVIDIA 3000-series GPU or newer. CPU execution is possible with enough free RAM, but it's much slower and may be unsuitable for interactive use.