How to re-use my custommade voices?

[helmingsmusik2-cmd]

Hello,
firstly thanks alot for a great little app for TTS generation. My wonder: How do I make use of my custom-made locally saved .pt voice? Im not a skilled coder so I might be missing smth basic...

Using Gordon, I somehow managed to generate and save my own .pt voicefile but I cannot get it to show up in the webUI.. (for reference Im using Blackwell 5070ti card)
What I have tried:

• docker run directly from YOUR page - but then I dont get the repo to upload my voice?
• docker compose (needed to chang dockerfile-optimized to 12.8.1 before building). I now have the repo an i found the voices folder where I added my custom.pt voice. When I run, the voice is still not showong.. do I need to rebuild?
• running the downloaded docker composition with following args in the prompt> docker run -p 8880:8880 -v "D:\my_custom_voice.pt" "kokoro-fastapi-gpu-kokoro-tts:latest" -It works well but my voice is not showing..

What am I missing; Do you have any detailed instructions on how to make this work for me?

Im mostly using Voice-Chat-AI from Bigsk1 with Kokoro-TTS and would like this to work together
Many thanks in advance//

[helmingsmusik2-cmd]

Update; I believe I got my custom voice to work by running it locally outside docker (astral UV?), however the TTS generation seems to be slower than inside Docker..

[michael-chipmates]

The voice list is just the .pt files the server finds in api/src/voices/v1_0/, and the voice name is the filename without .pt. So my_custom_voice.pt gets called as my_custom_voice.

You don't need to rebuild. The GPU compose already mounts your host api/ folder into the container:

volumes:
  - ../../api:/app/api

So put your file at api/src/voices/v1_0/my_custom_voice.pt on the host (make sure it's in v1_0/, not just voices/) and run docker compose restart. Then check the server actually sees it:

curl http://localhost:8880/v1/audio/voices

If my_custom_voice shows up in that list, you're good on the server side and the web UI just needs a reload. If it's missing, it's a path or permission thing. The container runs as UID 1001, so the file has to be readable by it.

Your docker run attempt didn't work because the -v had no target path. A bind mount is host:container, you only gave the host half. Point it at the file inside the voices dir so you keep the default voices too:

docker run -p 8880:8880 -v "D:\my_custom_voice.pt:/app/api/src/voices/v1_0/my_custom_voice.pt" kokoro-fastapi-gpu-kokoro-tts:latest

On the speed: outside Docker through uv it fell back to CPU. The default GPU build is cu126, which doesn't cover Blackwell, so your 5070ti sits idle. Uncomment the 50-series block in docker/gpu/docker-compose.yml and rebuild:

args:
  GPU_EXTRA: gpu-cu128
  CUDA_VERSION: 12.8.1

For Voice-Chat-AI, just set the Kokoro voice to my_custom_voice. It calls /v1/audio/speech with that name.