Phonon is a small text-to-speech system designed to run directly on phones, laptops, single-board computers, and web browsers. Because inference happens locally, typed text stays on the device and speech generation can continue without a network connection. The release follows six months of work with partners and early users after an April announcement.
The models are about 65 MB and 75 MB. Together, they cover English, French, German, Spanish, and Portuguese, with more than 100 built-in voices. Phonon 31m is an English model released under Apache 2.0, including commercial use. Phonon 40m covers all five languages under a non-commercial license. Voice cloning and voice design are also available through Gradium Studio.
Phonon uses flow matching and comes from the Pocket TTS lineage. Its dedicated ptts inference framework supports more than 10 backends across Intel and Arm CPUs, WebAssembly, CUDA, Vulkan, Metal, WebGPU, Apple Core ML, and Snapdragon NPUs. A reference PyTorch implementation is available in Transformers from version 5.21.
On Hugging Face’s Open TTS Leaderboard, Phonon ranked first for intelligibility in German and Spanish and placed in the top three for every evaluated language. It also led the CPU streaming benchmark for time to first audio. The models were downloaded more than 200,000 times during their first two weeks.