ZONOS2 is a leading open-weight text-to-speech (TTS) model designed for high expressiveness, quality, and naturalistic voice cloning. It targets engineers, researchers, and power users seeking to integrate advanced TTS capabilities, offering performance on par with or exceeding top commercial providers at low latency.

How It Works

ZONOS2 employs an open-weight Mixture-of-Experts (MoE) architecture, trained on over 6 million hours of multilingual speech. It leverages normalized UTF-8 bytes and ECAPA-TDNN embeddings to generate DAC tokens via its MoE backbone. This approach enables high-fidelity, naturalistic voice cloning and delivers expressiveness and quality comparable to leading TTS providers at low latency.

Quick Start & Requirements

• Platform Support: Linux only (x86_64).
• Prerequisites: NVIDIA GPU with compatible CUDA toolkit, uv package manager.
• Installation: Clone the repository (git clone https://github.com/Zyphra/Zonos2.git), navigate into the directory (cd Zonos2), and synchronize dependencies (uv sync).
• Launch TTS Server: Execute uv run python -m zonos2 --model-path Zyphra/ZONOS2 --tts-default-voices-dir ./default_voices/. The server is accessible at http://localhost:1919.
• Offline Inference: Utilize the TTSLLM class directly within Python scripts.
• Documentation/Demo: Hosted version available at cloud.zyphra.com/audio-playground. Speech samples and details on the blog (link not provided).

Highlighted Details

• Extensive Multilingual Support: Covers 3 tiers of languages, including English, Mandarin, Japanese, Korean, Russian, French, Spanish, and many others.
• High-Fidelity Voice Cloning: Achieves naturalistic and accurate voice cloning.
• Low-Latency Inference: Optimized for fast TTS generation.
• OpenAI-Compatible API: Provides a /v1/audio/speech endpoint mirroring OpenAI's API.
• Large-Scale Training: Trained on over 6 million hours of speech data.

Maintenance & Community

No specific details regarding contributors, sponsorships, or community channels (e.g., Discord, Slack) were found in the provided README.

Licensing & Compatibility

ZONOS2 is released under the permissive MIT License. It incorporates third-party components, including Mini-SGLang (MIT) and NVIDIA NeMo-text-processing (Apache-2.0), with details in the NOTICE and licenses/ directories. The MIT license generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

The system is restricted to Linux (x86_64) and requires NVIDIA GPU hardware with CUDA support for operation.