RELEASE · MODELS · #3130
Gradium launches Phonon — small on-device TTS models (31m Apache‑2.0; 40m CC BY‑NC)
Gradium released Phonon, a family of compact on-device text-to-speech models (about 65–75 MB) with two published sizes: Phonon 31m (Apache‑2.0) and Phonon 40m (CC BY‑NC). The models cover English, French, German, Spanish and Portuguese, include 100+ built-in voices and voice-cloning tools, ship with a new ptts inference framework and a Transformers reference in v5.21, and ranked first in German and Spanish on Hugging Face’s Open TTS Leaderboard while leading CPU time-to-first-audio benchmarks; the release saw 200,000+ downloads in two weeks.
KEY POINTS
- Gradium released Phonon, a family of compact on-device text-to-speech models (about 65–75 MB) with two published sizes: Phonon 31m (Apache‑2.0) and Phonon 40m (CC BY‑NC).
- The models cover English, French, German, Spanish and Portuguese, include 100+ built-in voices and voice-cloning tools, ship with a new ptts inference framework and a Transformers reference in v5.21, and ranked first in German and Spanish on Hugging Face’s Open TTS Leaderboard while leading CPU time-to-first-audio benchmarks; the release saw 200,000+ downloads in two weeks.
- On-device, open-weight TTS that matches much larger models and tops latency benchmarks can materially change privacy, deployment costs and latency for voice features across devices.
WHY IT MATTERS
On-device, open-weight TTS that matches much larger models and tops latency benchmarks can materially change privacy, deployment costs and latency for voice features across devices.