Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/IAHispano/Applio/llms.txt

Use this file to discover all available pages before exploring further.

Once Applio is installed, performing your first voice conversion takes only a few minutes. You need two things before you start: a source audio file (the voice you want to transform) and an RVC model (the target voice you want to convert to). Applio can download models directly from URLs through its built-in Download tab, so you do not need to manage files manually. This guide walks through the complete process using both the Gradio web UI and the command-line interface.

​
Model Files

Every RVC voice conversion requires two model files, both stored under logs/<model_name>/ inside your Applio directory:
FileExtensionPurpose
Model weights.pthThe trained neural network that encodes the target speaker’s voice
Feature index.indexA FAISS index of training embeddings used to refine the conversion
Both files are required. The index file controls how closely the output tracks the training data; if you only have a .pth file, set --index_rate 0 to skip index retrieval.

​
Path 1: Web UI

The web UI is the recommended starting point for new users. It exposes all inference parameters through interactive controls and provides an audio player for immediate preview of your output.
1

Start Applio

Launch Applio with the run script for your platform:
run-applio.bat
Applio starts a local Gradio server and opens your default browser at http://127.0.0.1:6969. Wait for the interface to fully load before proceeding — the first launch downloads prerequisite model files and may take a minute or two.
2

Download a voice model

Navigate to the Download tab. Paste a direct URL to a .pth file (or a .zip archive containing a .pth and .index pair) into the model link field, then click Download.Applio’s model_download_pipeline will fetch the files and place them in logs/<model_name>/ automatically. You can also upload model files manually to that directory.
Community-shared models are available on Hugging Face and the Applio website. Look for packages that include both a .pth and a .index file for best results.
3

Open the Inference tab

Click the Inference tab. In the Model dropdown, select the model you just downloaded. The dropdown is populated from the files present in logs/.
4

Upload your audio and configure parameters

Upload your source audio file using the input audio widget. Then review the key conversion parameters:
ParameterDefaultDescription
Pitch0Semitone shift applied to the output pitch. Range: −24 to +24. Use +12 (one octave up) when converting a male voice to a female model, or −12 when doing the reverse.
F0 MethodrmvpePitch extraction algorithm. rmvpe is the recommended default. Options: rmvpe, fcpe, crepe, crepe-tiny, and hybrid combinations such as hybrid[rmvpe+fcpe].
Index Rate0.3Influence of the .index file on the output (0.0–1.0). Higher values produce a closer match to the training voice but may introduce artifacts.
Volume Envelope1.0Blending ratio for the output volume envelope. 1.0 uses the converted audio’s envelope directly.
Protect0.33Protects consonants and breath sounds from conversion artifacts (0.0–0.5).
For singing voice conversions, enable F0 Autotune to snap the output pitch to the nearest chromatic note. Adjust Autotune Strength (0.0–1.0) to control how aggressively the pitch is quantized.
5

Convert and download your output

Click the Convert button. Applio will run the inference pipeline and display the output audio in the player below. Use the player controls to preview the result, then click the download icon to save the file.The output format defaults to WAV. To change it, select a different option from the Export Format dropdown before converting. Supported formats are: WAV, MP3, FLAC, OGG, M4A.

​
Path 2: CLI

The core.py CLI exposes the same inference pipeline as the web UI and is ideal for scripting, batch automation, or integrating Applio into larger workflows. Every parameter available in the UI has a corresponding CLI flag. The minimal command for a single-file conversion is: 
python core.py infer \
  --input_path audio/input.wav \
  --output_path audio/output.wav \
  --pth_path logs/MyModel/MyModel.pth \
  --index_path logs/MyModel/MyModel.index \
  --f0_method rmvpe \
  --pitch 0

​
Common CLI Parameters

FlagDefaultDescription
--input_path(required)Full path to the source audio file
--output_path(required)Full path where the converted file will be saved
--pth_path(required)Path to the .pth model weights file
--index_path(required)Path to the .index FAISS index file
--pitch0Semitone pitch shift (−24 to +24)
--f0_methodrmvpePitch extraction algorithm
--index_rate0.3Index file influence (0.0–1.0)
--volume_envelope1.0Output volume envelope blending
--protect0.33Consonant protection level (0.0–0.5)
--export_formatWAVOutput format: WAV, MP3, FLAC, OGG, M4A
--embedder_modelcontentvecSpeaker embedding model
--split_audioFalseSplit long audio into segments before inference
--clean_audioFalseApply noise reduction to the output
--f0_autotuneFalseSnap output pitch to the chromatic grid

​
Batch Inference

To convert an entire folder of audio files in one call, use the batch_infer subcommand: 
python core.py batch_infer \
  --input_folder audio/input_folder/ \
  --output_folder audio/output_folder/ \
  --pth_path logs/MyModel/MyModel.pth \
  --index_path logs/MyModel/MyModel.index \
  --f0_method rmvpe \
  --pitch 0 \
  --export_format FLAC

​
TTS-to-RVC Pipeline

You can synthesize text with edge-tts and immediately pipe the result through RVC conversion in a single command: 
python core.py tts \
  --tts_text "Hello, this is a test of Applio voice conversion." \
  --tts_file "" \
  --tts_voice en-US-AriaNeural \
  --tts_rate 0 \
  --output_tts_path audio/tts_out.wav \
  --output_rvc_path audio/rvc_out.wav \
  --pth_path logs/MyModel/MyModel.pth \
  --index_path logs/MyModel/MyModel.index

​
Output Format Reference

Applio supports the following output containers for all inference modes:
FormatFlag valueNotes
WAVWAVLossless, uncompressed. Default output format.
MP3MP3Lossy, widely compatible. Good for sharing.
FLACFLACLossless, compressed. Recommended for archiving.
OGGOGGLossy, open format.
M4AM4ALossy, AAC-based. Good for Apple ecosystem compatibility.

​
Next Steps

You have completed your first voice conversion. From here you can explore the full set of inference parameters, train your own RVC model, or extend Applio with plugins.

Full Inference Reference

Complete documentation for every inference parameter, post-processing effect, and formant-shifting option.

Training a Model

Learn how to preprocess a dataset, extract features, and train a custom RVC voice model from scratch.

Build docs developers (and LLMs) love

Get started for freeTalk to us