Hi all,
I wanted to try the recently released tts model from kyutai, which seems to be more efficient that kokoro and more reliable than supertonic for my hardware (ancient haswell CPU). I created a small server which uses an OpenAI compatible API to server the model in a docker environment (for instance). Works great with the OpenAI TTS custom component.
My repo can be found here https://github.com/bozakov/pocket_tts_api if you want to try it out.