A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.

Why it matters

Phonon TTS is a small, open text-to-speech model from Gradium that runs entirely on a user's own device, with no server or network connection required. Its English weights, released as Phonon 31m, carry a permissive Apache license that allows commercial use, while a larger sibling covering additional languages ships under a non-commercial license. The model draws on the Pocket TTS lineage and uses flow matching to reach quality previously associated with far larger systems, though its creators say training was initially numerically unstable and required special handling. On Hugging Face's Open TTS Leaderboard, Phonon TTS ranks first in German and Spanish intelligibility and places in the top three across every language evaluated, against models many times its size. It also tops the CPU streaming benchmark for time-to-first audio. With Hugging Face providing day-one Transformers integration and developers already shipping conversions and apps, the launch signals growing momentum for private, on-device speech synthesis.