Support for VoxCPM2?

[dougmaitelli]

Describe the Issue
Sorry if I am doing something wrong and this is not a bug, but I tried to run VoxCPM2 on KCPP and it failed to start because it identifies the model as OuteTTS and requires me to provide a tokenizer. VoxCPM2 is a tokenizer-free model.

Additional Information:
Running on a Strix Halo in Fedora 43. I am able to run other models on KCPP and other frontends without issues.

[henk717]

We don't want to integrate to many TTS backends. What makes this one substantially better than the Qwen we already have?

[henk717]

Also looked just in case its faster for example but there is no ggml implementation available for us to integrate.

[LostRuins]

As henk said, we recommend using Qwen3TTS or Kokoro, both of which are supported in KoboldCpp.

[dougmaitelli]

Yeah, I get it. I was mostly just trying to test and compare them because from what I saw so far Vox seems to have a bit more human result than Qwen.