Desktop real-time voice conversion using ONNX Runtime and RVC based models. Capture from a microphone, run inference locally (CPU or GPU), and hear converted audio on your chosen output device or on a locally saved file.
example.mp4
- Live : Capture audio from the selected input device, convert through your RVC model in real time, and play the result on the selected output device (with waveform monitor).
- Record : Record microphone input, then run offline conversion; output is written next to the executable as
temp_rvc.wav(raw capture astemp.wav). - Model selection : a voice ONNX from the
rvc_modelsfolder in the UI. - Processing backend : Auto, CPU-only, or GPU-preferred (CUDA).
- Pitch : Adjust semitones (−24 … +24)
| Component | Notes |
|---|---|
| OS | Windows (primary target; post-build steps copy DLLs beside the exe) |
| Qt | 6.x with Widgets, Multimedia, and Charts |
| Compiler | MSVC 2022 (64-bit) recommended |
| CMake | 3.16+ |
| ONNX Runtime | Prebuilt 1.25.0 with include/ and lib/ (see below) |
| GPU (optional) | Matching CUDA, cuDNN, and optionally TensorRT runtimes if you use the CUDA/TensorRT execution providers |
- Create
Externalfolder at the project root and add dependencies that match the versions inCMakeLists.txt
- Voice models are not included in this repository. Export them from the RVC ecosystem as ONNX.
- Clone RVC-Project/Retrieval-based-Voice-Conversion-WebUI
- Follow their install and training docs.
- Export using ONNX export tooling.
tools/export_onnx.py(replace the original with modified export_onnx) - Copy the resulting .onnx into rvc_models/ next to LocalRVC.exe.