November 5 edition Source published November 5, 2026.
Analysis
Speech synthesis does not have to visit a server. Gradium says its Phonon models run locally, without a network for inference.
The November 5 post explains an earlier release. Phonon 31m English is Apache 2.0; the 40m multilingual model has a non-commercial licence, so the editions are not interchangeable for commercial use.
Local speech moves network, data-handling and hardware questions to the product team. Gradium's performance claims are the authors' own, so pronunciation, speed and licence terms still need checking on intended hardware.
Source: A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks. · 2026-11-05 · Primary source context · attributed account
Why it matters
Local speech shifts deployment and licensing choices onto the product team.