@hironoflan Hiii, unfortunately, I think Intel Xe Graphics likely won’t handle real-time inference smoothly. RVC itself requires at least a mid-range dedicated GPU to run properly (about >RTX 2060?). Any program using deep learning needs sufficient VRAM (video memory). Integrated graphics share system RAM, which is significantly slower than dedicated VRAM, resulting in much lower performance — despite newer integrated GPUs might have improved absolute performance compared to older dedicated GPU.
If real-time voice conversion isn’t necessary for you, you can still use RVC1006 go-web.bat for non-real-time processing: upload a pre-recorded audio file and convert it afterward. Do not use older versions, as the latest release features the highly efficient rmvpe pitch extraction algorithm, which delivers far superior results at lower computational cost compared to previous methods.
The go-realtime-gui.bat is only for real time voice conversion, and the go-web.bat is for infer(convert) and training purpose. I see you mention missing go-web.bat? Maybe make your own and test if it works?
runtime\python.exe infer-web.py --pycmd runtime\python.exe --dml --port 7897
pause
As for VCClient, I used that once, but didn't get any better result compare to RVC one, so I haven't follow it's development.
If you really love the idea of real time voice change, I heard that DDSP might require less resource? I haven't tried this.
And maybe other proprietary software can yield a better result than RVC, but I mainly use RVC for making songs, So I just tried few and stopped. I suggest you DO NOT pay any money to any program that claim "support" RVC but run better, it just repacked RVC code inside I guess?
If you into voice clone, something like let a character read text for you, you can try IndexTTS
