Zyphra releases ZONOS3-Mini open-source text-to-speech model with voice cloning

1. 5:22 PM
   SGLang@sgl_projectOfficial

   SGLang-Omni adds day-0 support for Zonos3-Mini voice-cloning TTS

   AISGLang-Omni now runs Zyphra's Zonos3-Mini on day 0, according to SGLang. The 800M-parameter MoE text-to-speech model activates 80M parameters per token. On one H100, first audio streams back in 57 ms.

   
2. 5:07 PM
   Zyphra@ZyphraAIOfficial

   Zyphra releases ZONOS3-Mini, an Apache 2.0 on-device voice model

   AIZyphra releases ZONOS3-Mini weights under Apache 2.0 for on-device use in games, narration tools, and assistant voices. It supports browser, native, and server runtimes, with day-zero SGLang support. Zyphra says it offers supported fine-tuned or post-trained versions for customers who contact the company directly.

3. 5:07 PM
   Zyphra@ZyphraAIOfficial

   Zyphra introduces ZONOS3-Mini, a voice model for characters

   AIZyphra introduces ZONOS3-Mini, a voice model for giving characters a voice. The model is available through Zyphra Inference, a browser playground, and open weights on Hugging Face, with code on GitHub.

   
4. 5:07 PM
   Zyphra@ZyphraAIOfficial

   Zyphra's ZONOS3-Mini generates speech in real time using partial activation

   AIZyphra says ZONOS3-Mini activates only part of the model at each generation step. Its new audio codec reduces the tokens needed to represent speech from 774 to 150 per second, which the company says enables real-time generation across a wide range of hardware.

5. 5:07 PM
   Zyphra@ZyphraAIOfficial

   Zyphra reports ZONOS3-Mini speech generation on Pixel 9 and MacBook Air

   AIZyphra says its ZONOS3-Mini speech model runs locally on a Pixel 9 phone, with time to first audio under 800 ms. On an M3 MacBook Air in the browser, time to first audio is under 450 ms, and both devices generate speech in real time. Zyphra says it trained the model on NVIDIA H100s and serves it on H200s through Zyphra Inference.

   
6. 5:07 PM
   Zyphra@ZyphraAIOfficial

   Zyphra introduces open-source ZONOS3-Mini text-to-speech model

   AIZyphra introduces ZONOS3-Mini, an open-source text-to-speech model with voice cloning and word-level control. The company says the model runs on-device and in real time, including on a phone. Zyphra offers 20 hours of free audio to people who comment "ZONOS" within 2 days.

   

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.