01 What happened
Gradium has released Phonon, a compact text-to-speech model that runs entirely on the user's device without a server or internet connection. Available in two sizes (approximately 65 MB and 75 MB), it supports English, French, German, Spanish and Portuguese with more than 100 built-in voices, plus voice cloning and custom voice design through Gradium Studio.
02 Key details
- Phonon runs fully on-device — no server, no network, and input text stays on the user's device.
- Phonon 31m (English only) is released under Apache 2.0, permitting free use including in commercial products.
- Phonon 40m (all five languages) is CC BY-NC; commercial licensing is based on user count, not speech volume.
- On the Hugging Face Open TTS Leaderboard, Phonon ranks first in German and Spanish intelligibility and tops the CPU streaming benchmark for time-to-first audio.
- Available in 🤗 Transformers from v5.21 as a reference PyTorch implementation, with the Transformers integration created by the Hugging Face team. A dedicated ptts inference framework supports over 10 hardware backends.
03 Why it matters
Phonon addresses privacy and latency concerns of cloud-based TTS by keeping data on-device while matching or surpassing the quality of models tens of times larger.
04 Who it matters to
Voice application developers, resource-constrained device manufacturers and users who value privacy in speech synthesis.
Original source