Gradium open-sources Phonon: sub-50M on-device TTS that keeps talking offline
HeyAmit_ · x · 2026-10-20
Gradium officially open-sourced Phonon, a suite of on-device TTS models all under 50M parameters, claiming a new standard for naturalness and accuracy across languages with the broadest inference support to date; Phonon-31m is Apache 2.0.
Because it runs fully locally, it keeps speaking with Wi-Fi off — one demo shows it mocking a gamer in German. Official benchmarks put Phonon 31m's English WER at 1.70%, well ahead of Kokoro (3.50%) and other open peers.
Related event: Gradium Open-Sources Phonon: Sub-50M On-Device TTS Models(6 posts)→
More from Models
- Anthropic models share near-identical personality profiles, with Haiku 5.5 weakest — ArtificialAnlys · 2026-10-21
- Google models consistently high on Slop, low on Transparency in personality profiling — ArtificialAnlys · 2026-10-21
- Meta's Muse models split into sub-clusters, with two ranking highest on Slop yet — ArtificialAnlys · 2026-10-21
- OpenAI models cluster into three personality groups; GPT-4o stands out on Slop — ArtificialAnlys · 2026-10-21
- AI models cluster into three personality families that cross lab boundaries, finds AA — ArtificialAnlys · 2026-10-21
- AA image editing leaderboard v2.1: Krea 3 enters, GPT Image 2.5 max leads at Elo 1199 — ArtificialAnlys · 2026-10-21