Phonon TTS Model Launched for High-Quality On-Device Speech Generation

On-Device Speech Synthesis

Phonon is a new text-to-speech model designed to run directly on user devices. This architecture means it does not require a server or network connection, keeping user data local. The model's weights have been made public, with Phonon 31m released under an Apache 2.0 license, allowing free use in commercial products. Other models are available under a CC BY-NC license for non-commercial projects.

Achieving Quality in Small Footprints

A key development challenge for Phonon was to achieve high-quality speech output from a small model size, a combination previously difficult to attain. The model is based on the Pocket TTS lineage and utilizes flow matching for audio generation, a technique distinct from those used in other TTS systems. This approach enabled the model to deliver quality results despite its compact size, though it presented training challenges related to numerical stability.

Model Specifications and Features

Phonon is available in two sizes, approximately 65 MB and 75 MB. These models support English, French, German, Spanish, and Portuguese, and include over 100 built-in voices. Users can also create custom voices using Gradium Studio. The system is designed to handle complex text elements like phone numbers, dates, and codes accurately. It is compatible with various devices, including phones, laptops, single-board computers, and web browsers.