[Feature request] 1.7B Qwen3 support

[henk717]

Compared to the 1.7B model in another library I am getting less accurate voice cloning, I tried editing the script to convert 1.7B instead but was unable to run the model.
It would be nice if the 1.7B could also be supported, here are the repo links

https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base

[paoletto]

i've added it, but with no license for this project it's not possible to publish it

[henk717]

Please do publish it or at least submit it to our KoboldCpp fork. This project is MIT licensed, or at minimum the author gave us permission to use it as MIT. #4

[LostRuins]

i've ported it to koboldcpp

[henk717]

Technically that means that for me this is solved but ill leave the issue open for the upstream project.

[mbaudier]

Was it a big development? Would it be complicated to submit a PR to this repo? @paoletto @LostRuins

Or at least point to the related commits somewhere? (which could then be used as patches in a .deb package)

[henk717]

On the koboldcpp side its LostRuins/koboldcpp@4189508 koboldcpp itself exposes the TTS as an online api which can be used by other software.

Original version is here paoletto@87294f5