Taplio - Build a growth routine on LinkedIn
Mini toolsPhonon
By Gradium
Phonon is an on-device TTS model from Gradium, updated on 26-05-2026. The May 2026 version is about 100M parameters and is designed to run entirely on edge devices instead of relying on cloud inference, making it useful for offline voice agents, vehicles, remote equipment, healthcare, consumer hardware, and other latency-sensitive or privacy-sensitive products. Gradium reports that the update improved Seed-TTS English WER from 1.48% to 1.00%, raised speaker similarity from 56.37% to 59.51%, removed the earlier 100-token minimum input padding, and added int8 quantization with no perceived audio-quality loss. Phonon is still in private beta.
New
Video
Gen 4
Overview
Phonon is Gradium’s private-beta 100M-parameter on-device text-to-speech model for low-latency, offline, privacy-sensitive voice generation.
About Gradium
Gradium develops audio language models (ALMs) providing ultra-low latency text-to-speech, speech-to-text, and voice cloning capabilities via API. Spun out of French AI lab Kyutai, it builds voice AI infrastructure for developers and enterprise voice agents, targeting sub-200ms latency with human-like expressiveness and multilingual support.
Industry:
Artificial Intelligence
Company Size:
15
Location:
Paris, Île-de-France, FR
Website:
gradium.ai
Tools using Phonon
No tools found for this model yet.
Last updated: May 27, 2026
