Installing Inference Models

While you can create your own inference models in Applio (see Training a Voice Model), you may be able to find pre-existing models ready for inference. You can find such models here. These can be installed in three ways:

1. Using the model downloader
2. Uploading the files through Applio’s interface
3. Manually placing the model files

RVC models typically have two components: A PTH file (short for PyTorch archive) where all the actual model weights are, and a much smaller Index file. The Index file is optional, and can be left off even if it is provided. This file helps enforce the accent of the original speaker.

Using the model downloader

1. Obtain the URL of a ZIP file containing the PTH and Index files.
2. Go to the Download tab of Applio.
3. Paste the URL into the Download Model panel’s Download Link field.
4. Click the Download Model button.
5. Wait for the download to complete.

Uploading the files through Applio’s interface

1. If the files are in an archive such as a “zipped folder”, extract them.
2. Go to the Download tab of Applio.
3. Use the Drop Files field to upload the files, by clicking it and selecting them, or by dragging and dropping them to it.
4. Wait for the upload to complete.

Manually placing the model files

1. If the files are in an archive such as a “zipped folder”, extract them.
2. If the files do not extract to their own folder, create one named after the model and put them in it.
3. Move that folder to be inside Applio’s logs folder.

In all three cases, the end result should be that both the PTH and Index files are, for the example of a model named “John Doe Voice”, inside Applio/logs/JohnDoeVoice.