Gradium has released Phonon, a small text-to-speech model that runs entirely on the user's own device with no server connection or network needed. The company says text typed into the system stays on the device. Two model sizes, at roughly 65 MB and 75 MB, cover English, French, German, Spanish and Portuguese. They come with more than 100 built-in voices and support for custom voices through voice cloning and voice design. Phonon 31m is released under Apache 2.0, allowing free use including in commercial products. Phonon 40m covers all five languages under a non-commercial license. Phonon is built on the Pocket TTS lineage and uses flow matching to generate audio. The team says this approach allowed high quality in a small model but made training numerically unstable, requiring several techniques to reach good results across all five languages. On the Hugging Face Open TTS Leaderboard, which covers 23 languages and scores intelligibility, speed and naturalness, the company reports that Phonon ranks first in German and Spanish intelligibility and in the top three in every language it was evaluated on, competing with models up to 100 times larger. It also reports the fastest time-to-first audio on the leaderboard's CPU streaming benchmark. Gradium built a new inference framework, ptts, that supports more than 10 backends across CPU, GPU and NPU hardware, and Phonon is available in Hugging Face Transformers as its reference PyTorch implementation. The company says the model was downloaded more than 200,000 times in its first two weeks. For Australian businesses and professionals, the practical impact is limited unless they are building voice features. The Apache 2.0 license on Phonon 31m makes it a low-cost option for commercial products. Phonon 40m's non-commercial license means commercial use would require a separate arrangement, with pricing based on the number of users rather than the amount of speech generated. On-device processing may reduce privacy exposure under the Privacy Act 1988 because text does not leave the device, but organisations should still check how their apps handle voice data. Developers testing offline voice for apps, field tools or kiosks may find it worth trying. Anyone not building voice products will see no change from this release.