On-device TTS upstart Gradium Phonon beats models 100x its size on Open TTS Leaderboard
RemiCadene · x · 2026-11-05
Hugging Face folks are teaming up with Gradium to build the best open speech models. Gradium's new on-device TTS model Phonon posts standout numbers on the Open TTS Leaderboard:
- #1 intelligibility in German and Spanish
- Top-three intelligibility in every evaluated language
- Leading CPU streaming time-to-first-audio
The headline: a small on-device model competing with models up to 100x its size — and a case for evaluating TTS across intelligibility, throughput, and latency rather than one overall ranking.
Related event: Gradium's tiny open-source Phonon TTS beats models 100x larger(3 posts)→
More from Multimodal
- Google Releases Gemini Omni 1.2 Flash: Faster, Cheaper, Longer Video Clips — Hans-Wermhatt · 2026-11-06
- Google ships Gemini Omni Flash 1.2 with 15-second video and 30% lower pricing — OfficialLoganK · 2026-11-06
- DRAMA expression-editing model launches free on Hugging Face for video and image editing — SimplyAnnisa · 2026-11-06
- Steampunk Warrior AI Art Wows With Intricate Gear and Brass Detail — azed_ai · 2026-11-06
- No filming, all AI: Seedance 2.5 + Suno workflow creates lifelike Italian music video — azed_ai · 2026-11-06
- GPT Image 2.5 creates Algiers tea-drinker character, animated with Seedance 2.5 — azed_ai · 2026-11-06