Today for AI
HOT RADAR
opensourceHEAT 15.8°

AI CURATED HOT STORY · 11/10/2026

Zyphra Open-Sources 800M Speech Model for Real-Time Voice Cloning on Phones

1 reports archived1 independent sourcesupdated 11/10/2026, 23:44:17
Synthesis & Latest Updates
1 Sources Cross-Validated

Zyphra has open-sourced ZONOS3-Mini under Apache 2.0, a text-to-speech model built on a sparse Mixture-of-Experts design with 800M total parameters and roughly 100M active per token, paired with a new 15 Hz codec (ZAC-1) that cuts the number of frames to generate, enabling real-time operation on phones and laptops. On Zyphra's own ZTTS2-Eval it scores 66.2 on the Capability Index, ahead of Fish Audio S2 Pro (61.7) and MOSS-TTS v1.5 (60.2), with clean-set WER of 7.79 and 80 ms time to first audio on a single H100; reference-clip cloning, however, drops to 44.7 on the same index, and all results come from the vendor's own benchmark.

LATEST/Zyphra has open-sourced ZONOS3-Mini under Apache 2.0, a text-to-speech model built on a sparse Mixture-of-Experts design with 800M total parameters and roughly 100M active per token, paired with a new 15 Hz codec (ZAC-1) that cuts the number of frames to generate, enabling real-time operation on phones and laptops. On Zyphra's own ZTTS2-Eval it scores 66.2 on the Capability Index, ahead of Fish Audio S2 Pro (61.7) and MOSS-TTS v1.5 (60.2), with clean-set WER of 7.79 and 80 ms time to first audio on a single H100; reference-clip cloning, however, drops to 44.7 on the same index, and all results come from the vendor's own benchmark.

Hourly heat curve

5 fully observed hours
Now
15.8
Peak
17.211-11
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. Marktechpost (@Marktechpost)T2·72 pts

    Zyphra Open-Sources 800M Speech Model for Real-Time Voice Cloning on Phones

    • 800M total parameters with only ~100M active per token makes sparsity the core enabler of on-device real-time inference
    • The in-house 15 Hz codec ZAC-1 reduces generated frames, reaching 80 ms time to first audio on a single H100
    • Clean-set WER of 7.79 beats Fish S2 Pro's 11.06, but reference-clip cloning scores only 44.7 and all results are vendor-run benchmarks