Player not loading? Watch on YouTube
This tutorial covers manual Windows installation of Applio 3.6.4 and voice conversion with a downloaded RVC model. The speaker says training can work with 3 to 6 GB of VRAM and that the resulting models can run in other RVC-compatible tools. Those are introductory claims; the demonstrated workflow focuses on setup and audio conversion. Training and voice calling are deferred to a later video.
The setup lists Python 3.10.11, CUDA Toolkit 11.8, FFmpeg, Visual Studio and Git for Windows. After cloning the repository, the speaker creates and activates a Python environment, installs PyTorch separately, then edits the requirements file. The specified changes include SciPy 1.15.1 and NumPy 1.26.4, plus removing torch and torch audio from that list. The speaker reports errors with the one-click installer and recommends manual installation based on that experience.
The first launch downloads models and provides a local browser URL. A run.bat file makes subsequent launches easier. For the local AI voice conversion example, the speaker downloads an RVC model and its index file, refreshes the inference view, selects the model and uploads an audio recording. After accepting the agreement and reviewing optional advanced settings, the speaker converts the recording and plays the result.
