Gradium Open-Sources Phonon: 31M-Parameter On-Device TTS Model Under Apache 2.0

Gradium has released Phonon, a 31-million-parameter text-to-speech model open-sourced under the permissive Apache 2.0 license, designed for efficient on-device inference. At 31M parameters, Phonon is small enough to run on mobile and edge devices without requiring cloud inference, making it relevant for offline-capable voice applications, accessibility tools, and privacy-sensitive deployments. The Apache 2.0 license removes commercial use friction, allowing developers to ship Phonon-powered products without licensing concerns. For teams building voice interfaces or TTS pipelines that need to avoid per-request API costs or latency, this is a compelling open-weight option to benchmark against existing solutions like Kokoro or Piper. The combination of size, license, and on-device target makes Phonon a practically useful addition to the open TTS ecosystem.