Gradium Releases Phonon: A Lightweight 31M-Parameter On-Device TTS Model
Gradium has open-sourced Phonon, a compact 31-million-parameter text-to-speech model available under the Apache 2.0 license. Requiring just 64.5 MB of storage and achieving a latency of around 7 milliseconds on standard laptop CPUs, this efficient on-device model enables fast, private, and accessible speech synthesis for developers and edge applications.
Source: MarkTechPost