AI Briefing
KOSign in

Gradium Open-Sources Phonon: 31M and 40M Parameter TTS Models for Edge Devices

·2026.10.21 01:01

Key point

The inference stack targets 10+ backends including CPUs, GPUs, and NPUs, with most components written in Rust.

Details

  • Gradium released Phonon, its first open-source model collection consisting of small Text-to-Speech models designed for on-device execution.
  • The model collection includes variants with 31M and 40M parameters.
  • The fully open-source inference stack supports 10+ different backends, covering CPUs, GPUs, and NPUs.
  • Most of the inference stack code is written in Rust.
  • The project is available on Hugging Face, with the company inviting community feedback and building on the codebase.
open-sourcetext-to-speechedge-airustgradiumphonon
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

RedditKyutai releases Pocket TTS Small models with 1.9M to 18.2M parameters for CPU voice cloningHacker NewsRespan releases open-source Rust AI gateway with plugin-based architectureRedditSteerTTS: 23M-parameter English TTS model with 12 numeric controls and real-time CPU inferenceHacker NewsCapsem: Open-Source Rust Hypervisor Sandboxes AI Coding Agents in Air-Gapped VMsGeekNewsArtCraft releases open-source IDE for AI image and video creation with 62-model catalogDeepSeekDeepSeek Releases V4.1-Flash-1017 and Opens Search & Fetch APIs