Gradium Releases Phonon: Open-Source 31M and 40M TTS Models for Edge
Gradium has open-sourced Phonon, a family of Text-to-Speech models featuring 31 million and 40 million parameters designed specifically for on-device deployment. The release includes a fully open-source, Rust-based inference stack supporting over 10 different hardware backends. This release provides developers with highly efficient, lightweight voice generation tools that can operate locally on resource-constrained hardware without cloud dependency. It lowers the barrier for integrating natural speech synthesis into mobile apps, web browsers, and edge devices. The Phonon models can stream 24 kHz audio while still generating text, and they are compact enough to run smoothly in a browser tab, on a smartphone, or on a laptop CPU. The underlying inference stack targets a wide variety of hardware options including CPUs, GPUs, and NPUs.
## BACKGROUND
Edge AI refers to running machine learning models locally on user devices rather than on remote cloud servers, which improves privacy, reduces latency, and works offline. Text-to-Speech (TTS) models translate written text into spoken audio, and optimizing them for edge devices requires shrinking model sizes while maintaining natural-sounding voice quality.