- Gradium released Phonon, a text-to-speech model of about 65–75 MB that runs on phones, laptops and in a browser.
- It speaks 5 languages, has more than 100 voices and was downloaded over 200,000 times in two weeks.
In short: Gradium released Phonon, a small text-to-speech model that runs entirely on your device, with no server and no network. The launch post appeared on the Hugging Face blog on 5 November 2026, two weeks after the model went out.
What you get
- Two sizes, about 65 MB and 75 MB. Between them they speak English, French, German, Spanish and Portuguese.
- More than 100 built-in voices, plus proper handling of phone numbers, dates and codes, which trip up many similar systems.
- It runs on phones, laptops, single-board computers and in a browser tab. In the launch demos Phonon talked a user through brewing coffee on an Apple Watch with Wi-Fi off and counted reps on an iPhone with no signal.
- You can make your own voice through voice cloning from a short recording (with the speaker’s consent) or by describing it in words in Gradium Studio.
- Gradium says the model was downloaded more than 200,000 times in its first two weeks.
What it costs
The smaller Phonon 31m speaks English only and uses the Apache 2.0 licence, so you can build it into a paid product for free. The five-language Phonon 40m is licensed for non-commercial use only (CC BY-NC); in a commercial product you pay by number of users, not by the amount of speech generated. Gradium trains extra languages on request.
How to read this
The results come from the company itself: Gradium writes that on Hugging Face’s Open TTS leaderboard Phonon ranks first for intelligibility in German and Spanish and in the top three in every language tested, next to models up to 100 times its size. The leaderboard is new, so listen to the samples yourself (it has a tab for side-by-side comparison). I write more about voice tools in the Image, video & voice hub, and about models that run on your own hardware in the Local AI hub.