A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.
Blog post from Hugging Face
Phonon is an open, on-device text-to-speech model family from Gradium designed to deliver high-quality speech without requiring a server or network connection, keeping user text local. Available in 65 MB and 75 MB versions, it supports English, French, German, Spanish, and Portuguese, includes more than 100 voices, offers voice cloning and voice design tools, and is intended for devices ranging from phones and browsers to laptops and single-board computers. The English-only 31-million-parameter model is released under Apache 2.0 for commercial use, while the multilingual 40-million-parameter version is non-commercial. Built on the Pocket TTS lineage using flow matching, Phonon reportedly overcame training-stability challenges to achieve strong quality at a small size, ranking first for intelligibility in German and Spanish and leading CPU streaming tests for time-to-first-audio on Hugging Face’s Open TTS Leaderboard. Gradium provides the ptts inference framework across CPU, GPU, browser, and NPU backends, while Hugging Face Transformers supports a PyTorch implementation for high-throughput generation and customization. Initial demonstrations included offline Apple Watch, iPhone, browser, robot, and game applications, and the project received more than 200,000 downloads in its first two weeks; future plans include custom model sizes, additional language support, vocabulary fine-tuning, and on-device voices created with consent.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.