According to a blog post published on Hugging Face on November 5, 2026, Gradium released Phonon, a text-to-speech (TTS) system designed to run on a user's own device without a network connection. The post states Phonon comes in two sizes, roughly 65 MB and 75 MB, and speaks English, French, German, Spanish and Portuguese, with more than 100 built-in voices and support for custom voices through voice cloning and voice design. According to the post, the smaller Phonon 31m model is English-only and released under the Apache 2.0 license, permitting commercial use, while the Phonon 40m model covers all five languages under a non-commercial (CC BY-NC) license. The company says Phonon uses a flow-matching technique from the Pocket TTS lineage and required training adjustments to remain numerically stable. On Hugging Face's Open TTS Leaderboard, which evaluates models across 23 languages on intelligibility, speed and naturalness, the post states Phonon ranked first in German and Spanish on intelligibility and in the top three in every language tested, and topped the CPU streaming benchmark on time-to-first audio. Gradium reports Phonon was downloaded more than 200,000 times in its first two weeks. The company built a dedicated inference framework, ptts, and the model is available in Hugging Face Transformers from v5.21. These figures come from the vendor's own announcement.
- Two models, ~65 MB and ~75 MB, covering English, French, German, Spanish and Portuguese
- Phonon 31m is Apache 2.0 (commercial use allowed); Phonon 40m is non-commercial (CC BY-NC)
- Runs entirely on-device with no server or network; text stays local
- Vendor reports first in German and Spanish on intelligibility and top of CPU time-to-first-audio benchmark
- Downloaded more than 200,000 times in its first two weeks, per Gradium
What it means for you
A text-to-speech model that runs on a phone, laptop or browser without sending anything to a server now sounds good enough to compete with much larger cloud services. The practical point for operations is privacy and no network dependency: spoken output can work offline, and the text being read aloud never leaves the device. For most organizations this is background knowledge, not a to-do — unless you already generate speech (IVR prompts, accessibility readouts, voice-enabled products) and worry about the cost or privacy of sending that text to a cloud vendor.
Try this
If any of your systems already turn text into speech, write down one thing: does that text currently leave your network to a third-party service, and does it ever contain customer or sensitive data? That single check tells you whether an on-device option like this is even worth a technical conversation.
Who should care
Teams building voice features into products, anyone running accessibility text-to-speech, and developers who need speech generation to work offline or keep text on-device for privacy reasons.
Skip this if
You don't generate speech from text in any workflow, or your current cloud voice setup works fine and sends no sensitive data. This is a developer tool, not something an operations team acts on directly.
Sources: Hugging Face — read the original