Gradium open-sources edge TTS engine Phonon

On October 23, 2026, MarkTechPost reported that Gradium had open-sourced Phonon, a family of on-device text-to-speech models. The English edition, Phonon 31m, is released under Apache 2.0, fits in 64.5 MB, and starts speaking in about 7 ms on a laptop CPU. The report described the series as built for local, on-device operation, though it offered no benchmarks comparing Phonon against other models.

The following day, October 24, 2026, DEV Community's AI section confirmed the open-sourcing of Phonon as Gradium's on-device TTS engine, pointing readers to gradium.ai/phonon. The author argued that the release matters less for the model itself than for what it signals: on-device speech synthesis is now a dependency teams can ship in production rather than treat as a research demo. The piece also cautioned that the hard part of edge speech is not the model. Audio is the one component in an application with a hard real-time deadline, the author noted: once playback starts, the synthesis loop must keep pace with DAC sampling continuously, or buffer underruns will produce clicks and robotic stuttering. As of these reports, Gradium has not published further details or follow-up statements on the project.