Gradium open-sources Phonon: 64MB on-device TTS with ~7ms first audio, 37 voices
lmoroney · x · 2026-10-22
Gradium has open-sourced Phonon, a family of tiny text-to-speech models built to run entirely on your own hardware, keeping voice data on-device. The English 31M-parameter model is about 64 MB, Apache 2.0 licensed, and ships with 37 voices; Gradium claims first audio in roughly 7 ms on a MacBook Air. The poster wrote up what's in the release, how it works, and which models fit which use cases.
Related event: Gradium Open-Sources Phonon, Sub-50M On-Device TTS Models(13 posts)→
More from Multimodal
- Entire Halloween Brand Video Made With AI: CapCut Image+Video+Audio on One Timeline — JaynitMakwana · 2026-10-22
- Kling 4.0 Lands on Magnific: 30-Second Scenes, 10 Keyframes, Native Audio — aziz4ai · 2026-10-22
- Director says AI cut his film's budget from $80M to $1.5M, but distribution remains the bottleneck — DavidmComfort · 2026-10-22
- Creator Uses iPhone Footage as Reference, AI Rebuilds It Into Motion Graphics Worlds — aziz4ai · 2026-10-22
- Kling 4.0 API officially rolls out with 30-second videos and multi-reference inputs — aziz4ai · 2026-10-22
- Seed Audio 1.5 teased: one prompt generates dialogue, SFX and music as separate tracks — heyronir · 2026-10-22