help with CosyVoice 3 en VoiceStudio, doesn´t work!

[JCisHEREpe]

CosyVoice 3 appears as Installed in VoiceStudio, and the model is already downloaded at C:\Users\xxxx\AppData\Local\OmniVoice\hf_cache\models--FunAudioLLM--Fun-CosyVoice3-0.5B-2512.

However, when I select CosyVoice 3 as the TTS engine, VoiceStudio shows:
Engine unavailable. Check installation and configuration.

When I click “What is needed?”, VoiceStudio shows these requirements:

git clone --recursive FunAudioLLM/CosyVoice
pip install -r requirements.txt
SoX

I’m using Windows 11 with an RTX 3050 8GB. Other TTS engines work correctly.

What exactly is missing or incorrectly configured, and what is the correct way to install/fix these requirements so CosyVoice 3 works directly inside VoiceStudio?

[basil-k-aji-dev]

The downloaded weights are not what the availability check looks at, which is why it reads as installed in one place and unavailable in the other.

CosyVoiceSubprocessBackend.is_available() (backend/engines/cosyvoice_subprocess/__init__.py) only asks whether the engine has its own virtualenv:

@classmethod
def is_available(cls) -> tuple[bool, str]:
    if own_venv_python() is None:
        return False, (
            "cosyvoice package not installed. Install it from "
            "Model Catalogue → Engines."
        )
    return True, "ready"

CosyVoice runs in a killable sidecar with its own interpreter, separate from the app's. Your hf_cache/models--FunAudioLLM--Fun-CosyVoice3-0.5B-2512 is the model weights — a different thing, and the check never looks there.

own_venv_python() resolves through engine_venv_python("OMNIVOICE_COSYVOICE_DIR"), which returns a path only when three conditions hold:

1. OMNIVOICE_COSYVOICE_DIR is set
2. <that dir>/.venv/Scripts/python.exe exists (Windows layout)
3. <that dir>/.voicestudio_install_complete exists

The third is the one that usually bites. From the source:

The interpreter alone proves nothing: a reinstall that failed partway leaves it behind. The completion marker is written only after the engine's import probe passed in this venv, and removed when a new dependency step starts.

So a half-finished install leaves a real-looking .venv on disk with no marker, and the engine stays unavailable no matter how many times you restart. That matches your symptom exactly.

What to do: install CosyVoice from Model Catalogue → Engines inside VoiceStudio, not by hand. That one-click installer creates the venv, installs the requirements into it, runs the import probe, and only then writes the marker.

To confirm which condition is failing, check whether the marker exists. If .venv is there but .voicestudio_install_complete is not, the install died partway — reinstall from Model Catalogue and watch it through to completion rather than restarting the app.

On the git clone --recursive / pip install -r requirements.txt / SoX list: that is upstream CosyVoice's own setup documentation. You do not need to run it yourself for the in-app engine — the managed installer handles the equivalent inside the engine's venv. Installing those into your system Python will not make is_available() return true, because it is looking at OMNIVOICE_COSYVOICE_DIR, not your global environment.

Same design applies to the other sidecar engines (VoxCPM2, MOSS, IndexTTS), so if one of those ever reports the same thing, it is the same marker check.