Hexgradai lab
Hexgrad
This is a Model Atlas entity, not a ranked standalone lab. A plausible organization withheld from the directory until independent evidence clears the publication gate.
Kokoro showed that a very small speech model could achieve useful quality and broad developer adoption without proprietary weights.
Profile
Overview
Hexgrad is the publisher of Kokoro, an 82-million-parameter open text-to-speech model built on the StyleTTS 2 architecture. Kokoro is Apache licensed and designed for fast, economical local or hosted speech synthesis.
The model was trained by the contributor known as rzvzn using permissively licensed and public-domain audio, with architecture work credited to the StyleTTS 2 team. Hexgrad operates as a small open-model project rather than a conventionally documented company.
Notable contributions
- 01Compact open speech synthesisKokoro delivered practical multilingual text-to-speech in an 82-million-parameter checkpoint with permissive licensing.