AI Briefing
KOSign in

Zyphra Releases ZONOS3-Mini, an Open-Source 100M Parameter TTS Model

·2026.11.11 16:09

Key point

The model supports 43 languages and runs on phones, laptops, and browsers using Apache 2.0 licensed weights.

Details

  • Zyphra released ZONOS3-Mini, a compact open-source text-to-speech model with 100 million active parameters that enables real-time inference on phones, laptops, and directly in the browser.
  • The model supports 43 languages and allows voice cloning from audio samples or character design by shaping age, accent, and vocal texture.
  • Architecture utilizes a latent Mixture of Experts (MoE) backbone with 64 experts and top-2 routing, generating 15 Hz dual-codec frames rendered to 44.1 kHz audio via a causal multi-resolution iSTFT head.
  • Weights are released under Apache 2.0, with native runtimes available for NVIDIA GPUs (CUDA), Apple Silicon (MLX), CPU/GPU (C++/ggml), and WebGPU in-browser builds.
  • Inference controls include loudness, bandwidth, and speaking rate, with a context window of 32,768 positions and support for up to 10 minutes of audio per generation.
open-sourcetext-to-speechzyphravoice-cloningreal-time-inferencezonos3-mini
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditKittenML Releases Kitten TTS 2 Mini, Nano, and Pico Models for CPU InferenceRedditTontaube V1.1 released: Open-weights TTS with 4-bit CPU support and 70% preference over Fish Audio S2 ProElevenLabsOpen Source TTS vs Paid APIs: Tradeoffs for Control, Cost, and Data SovereigntyPyTorchKRllama.cpp v1.0.0, First Major Release in Three and a Half YearsGeekNewsGitHub Next releases Ace, a local collaborative coding environment for humans and AI agentsCloudflare BlogCloudflare open-sources Kitesurf, its stateless browser for AI agents