AI BriefingGradium Details Phonon TTS: On-Device Speech Model Tops CPU Streaming Benchmarks
·2026.11.05 09:00
Key point
Gradium's Phonon TTS, available in 31m and 40m sizes, tops the Hugging Face Open TTS Leaderboard's CPU streaming benchmark for time-to-first audio.
Details
- Gradium published a detailed look at Phonon, an open-weight text-to-speech model launched two weeks prior, available in 31m and 40m parameter sizes, designed to run entirely on-device without network access.
- Phonon 31m achieves a 95th-percentile time-to-first audio of 28 ms, ranking first on the Hugging Face Open TTS Leaderboard CPU streaming benchmark, ahead of Pocket TTS at 165 ms.
- The model ranks first in intelligibility for German and Spanish and places in the top three for all evaluated languages, competing with models up to 100 times its size.
- Phonon 31m speaks English and is licensed under Apache 2.0 for commercial use, while the 40m model covers five languages (English, French, German, Spanish, Portuguese) under a non-commercial license.
- The release includes a new inference framework, ptts, supporting over 10 backends including CPU, GPU, and NPU, and is integrated into Hugging Face Transformers v5.21.
- The model uses flow matching from the Pocket TTS lineage and supports voice cloning and design via precomputed speaker latents.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.