| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| Applio 3.3.0 source code.tar.gz | 2025-08-18 | 13.3 MB | |
| Applio 3.3.0 source code.zip | 2025-08-18 | 13.5 MB | |
| README.md | 2025-08-18 | 1.9 kB | |
| Totals: 3 Items | 26.8 MB | 0 | |
Important Installation Guidelines
### Before You Install Applio
Please review these essential instructions to ensure a smooth installation:
- File Location: Place the Applio files in a dedicated folder on your C drive.
- Access Control: Avoid placing Applio in folders that require elevated privileges.
- Administrator Rights: Do not run the script with administrator privileges.
- Path Requirements: Ensure the installation path is free of spaces or special characters.
- Security Settings: Temporarily disable your antivirus and firewall to prevent any issues with missing dependencies during installation.
### Installation Steps
- Download Compiled Version: Get the zip from this link.
- Run Applio: Extract the file, then open the run-applio.bat file.
What's New in This Release
- Added support for Python 3.11
- Updated libraries, including a fix for EdgeTTS (without limits or errors for now)
- Ensured compatibility with RTX 50-series GPUs
- Reworked f0 predictor (removed unnecessary hop length, RMVPE now supports longer audio inputs)
- Added warnings when training steps are skipped
- Enhanced TensorBoard reference system (users can now create custom references)
- Improved discriminators and losses (renamed
d/total→d/advfor consistency withg/adv) - Introduced proposed pitch
- Set single-scale mel loss as the default loss
- Fixed audio file extension checks to be case-insensitive
- Updated speaker count logic for model training
- Added BFloat16 (Nvidia Ampere cards or newer) and FP16 precision for training. FP16 is much faster on Kaggle and Google Colab T4s free GPUs.
Full Changelog: https://github.com/IAHispano/Applio/compare/3.2.9...3.3.0