Physix FrontierPhysix Frontier · News Briefing Card (HuggingFace · Nov 5, 2026)
Gradium launches on-device TTS model Phonon, 200K+ downloads in first two weeks
KEY FACTS
- Gradium has released Phonon, an on-device speech synthesis model that runs without an internet connection or servers.
- Phonon comes in two sizes, roughly 65MB and 75MB, and supports five languages in total.
- Phonon 31m is released under the Apache 2.0 license and can be used commercially for free; the other models carry non-commercial licenses.
- On the Hugging Face open TTS leaderboard, Phonon ranks first in German and Spanish intelligibility.
- Phonon surpassed 200,000 downloads within two weeks of release and is still on the Hugging Face trending list.
KEY DATA
~65MB / 75MBModel size
5Number of supported languages
100+Number of built-in voices
200K+Downloads in first two weeks
PHYSIX OBSERVATION
The competitive focus in on-device TTS is shifting from whether it can run at all to how fast and how accurately it runs. With its tiny footprint, Phonon leads by a wide margin on first-token latency, showing that the experience bar for local voice interaction has been lowered. For users, privacy and offline availability are tangible benefits; for the industry, open weights plus a commercial license will accelerate the embedding of voice capabilities into all kinds of hardware.
Source: HuggingFace report