Skip to content
Hot eventPast event

Zyphra releases open-source speech model ZONOS3-Mini

2 reports1 sourceUpdated 3 days ago
Subscribe to this event via RSS

Start here

AI overview

Zyphra releases open-source speech model ZONOS3-Mini, with about 100 million active parameters, supporting 43 languages, and can run in real time on phones, laptops, and browsers. The model supports voice cloning from audio samples, designing characters by age/accent/timbre, and word-by-word control of emotion, speed, and expression. The model weights are open-sourced under Apache 2.0, and can be served via SGLang-Omni on NVIDIA GPUs with an OpenAI-compatible interface; browser execution requires WebGPU support. The official also provides a native transformers port, using ZAC-1 codec as audio tokenizer, with 27 built-in release voices.

Written by AI from the reports · updated 3 days ago

Developments

2 developments

  1. Nov 16 10:53 · 1 report
    Zyphra releases native transformers port of ZONOS3-Mini
    Zyphra · Zonos 语音模型:Zyphra releases ZONOS3-Mini multilingual TTS model
  2. Nov 9 13:03 · 1 report
    ZONOS3-Mini open-source speech model released
    Zyphra · Zonos 语音模型:Zyphra releases open-source speech model ZONOS3-Mini

Report timeline

Follow the reports to see the event from other sides.

Nov 16
  1. Zyphra · Zonos 语音模型Picks
    Zyphra releases ZONOS3-Mini multilingual TTS model

    Zyphra releases compact multilingual TTS model ZONOS3-Mini, supporting 43 languages, with approximately 100 million active parameters per token, and provides a native transformers port. The model uses ZAC-1 codec as audio tokenizer, includes 27 built-in release voices, supports voice cloning from short audio, and is licensed under Apache-2.0.

Nov 9
  1. Zyphra · Zonos 语音模型Picks
    Zyphra releases open-source speech model ZONOS3-Mini

    Zyphra releases open-source compact speech model ZONOS3-Mini, with approximately 100M active parameters, capable of real-time operation on phones, laptops, and browsers. The model supports cloning voices from audio samples, designing characters by age/accent/timbre, and word-by-word control of emotion, speaking rate, and expression. Weights are open-sourced under Apache 2.0, and it can be served via SGLang-Omni on NVIDIA GPUs with an OpenAI-compatible interface.

Heat trend for this event

Current
3
Peak
10Nov 16 12:00
24h change
-49%

The trend compares only sources observed the whole time, so the range may be smaller than the current heat figure. Hover or tap the chart for each hour. Left and right arrow keys move through it.