https://github.com/roryeckel/wyoming_openai
I created this project a while back and it's pretty stable. It's my favorite for using Kokoro with Home Assistant, but I plan to provide examples for future TTS/STT technologies as they are released.
It's why I decided to create this proxy server, because future text-to-speech and speech-to-text instantly become compatible with Home Assistant simply if they are OpenAI-compatible. No need to install an integration inside Home Assistant.
If Wyoming were directly tied to Kokoro, then this opportunity would not be possible.
The project is used primarily through docker. It can also make use of GPUs, unlike the Piper and Whisper integrations from Home Assistant.
Plus, if you use OpenAI somewhere else in your lab, you can consolidate your resources and avoid running two stacks.
Enjoy!