I wrote a guide here and show how to connect Kokoro FastAPI with Open WebUI

[ItsNoted]

Learn how to get Kokoro FastAPI setup on your own hardware and connect it to your Open WebUI instance!

https://noted.lol/kokoro-fastapi/

[RBEmerson970]

Regarding setting the Admin > Audio for TTS using Dockerized Kokoro-FastAPI, please see the attached grab.

[m-y-n-o-n-a]

Can you extend your tutorial and include also how to use gpu and not only cpu? That would be very helpful. Thx

[RBEmerson970]

@m-y-n-o-n-a
Read the README- the quickest way to a Docker build is:
docker run --gpus all -p 127.0.0.1:8880:8880 ghcr.io/remsky/kokoro-fastapi-gpu:latest
Note I added a specifical localhost IP as defaulting to 0.0.0.0 doesn't work for me, and 127.0.0.1 should work for most cases.

This is the only decision point between cpu and gpu support. Open WebUI doesn't directly care about cpu vs gpu for speaking messages. (In the case of thinking models, Open WebUI will not speak that part, only the final results.)