Different models for tts/stt

[pallid sage]

I was wondering if it's possible to use anything besides piper and whisper.

[brazen patrol]

there are some projects that use other systems and add a Wyoming wrapper to them so they can be integrated.
is there something in particular you are looking to use?

[pallid sage]

I wanted to try the new qwen3-tts model

[brazen patrol]

Interesting, I am sure its possible to write a wyoming wrapper on top of it and integrate it. nobody has done it yet though as far as I know.

I suspect someone will do it. but it takes quite a bit of compute to run it.

Actually a pretty cool idea if you have the VRAM to run it on something.

[pallid sage]

Makes sense, I'm running all my TTS/stt/llm on my workstation ATM and connecting ha to it from the rpi

Yeah I'm running on 7900xtx

[brazen patrol]

It looks like the demo project for running it locally is for cuda cards.

[pallid sage]

Where is that?

[brazen patrol]

pretty good write up here - https://dev.to/czmilo/qwen3-tts-the-complete-2026-guide-to-open-source-voice-cloning-and-ai-speech-generation-1in6

[pallid sage]

Interesting, I figured I can run it with llama-server

[brazen patrol]

I am sure its "possible" to run the model on non-cuda cards with the standard tools. however might take some time before all the pieces are in place. remember its only been a few days.

[pallid sage]

Torch and stuff has AMD/vulkan versions so it wouldn't be a problem really

But I was wondering about any Wyoming proxy kinda thing

[brazen patrol]

I think its definitely doable, just need someone with a lot more knowledge then me to put it all together.

[pallid sage]

I'll try to dig more into it

[brazen patrol]

Wyoming spec is available HERE so it should be doable to make a wrapper for something running it.

Or if someone made an OpenAI-TTS wrapper you could probably connect it using THIS custom integration.

ooo, looks like someone has made an OpenAI-TTS wrapper HERE.
If you can get that running you can use the above custom integration with it I imagine.

[pallid sage]

Ooohhh that looks interesting

[brazen patrol]

I am sure someone will make a wyoming wrapper for it soon enough tbh. but maybe for now can mess with using the openai wrapper with the custom endpoint integration.

Just a matter of trying to fit the pieces together.

[pallid sage]

Man this is a rabbit hole lol

[brazen patrol]

Yeah, thats kinda the nature of this type of thing 😛

[dapper wren]

For what it's worth, you can run Kokoro TTS which is quite good and runs a very fast on my 3070 https://github.com/remsky/Kokoro-FastAPI

connects with the OpenAI TTS integration

[pallid sage]

Thanks!