Back to feed
Zonos3-Mini: Speech Generation and Voice Cloning in the Browser

Zonos3-Mini: Speech Generation and Voice Cloning in the Browser

Zyphra has released Zonos3-Mini, an open-source model designed for high-precision voice cloning and speech synthesis. With just 100 million parameters, it runs in real-time on smartphones and laptops directly in the browser using WebGPU technology.

Voice Characteristic Control

Despite its compact size, Zonos3-Mini offers granular acoustic control. Users can dynamically adjust speaker age, accent, timbre, and emotional tone. The system allows control over speech tempo and intonational stress down to individual words, ensuring natural sound without the typical robotic artifacts.

Multilingualism and Cross-Platform Compatibility

The model architecture supports 43 languages out of the box, including Russian, English, Spanish, Chinese, Japanese, and Arabic. A key advantage of this release is full autonomy: Zonos3-Mini compiles to GGUF and Safetensors standards and executes natively on local GPUs via WebGPU. This eliminates the need to install Python, CUDA, or call server-side APIs.

Accessibility and Integration

Open model weights are available on Hugging Face, while source code and web runtime are published in the project repository. For high-load production systems, engineers have access to serverless inference via Zyphra Cloud. The Apache 2.0 license allows free use of checkpoints and ready-made quantizations.

6.1K views

More from this channel Metavershche and AIshche

Similar in this category Technology