Created with the purpose for curating datasets for the sake of training AI models. This is created with RVC (Retrieval-based Voice Conversion) in mind but generally works for any other AI voice model that needs short clips less than 10s.
- Python 3.10 installation
- git installation
- vscode installation (highly recommended)
- ffmpeg installation
- Cuda Capable Nvidia GPU (highly recommended)
- Clone the repository (repo)
git clone https://github.com/JarodMica/audiosplitter_whisper.git
- Navigate into the repo with:
cd audiosplitter_whisper
- Run setup-cuda.py if you have a compatible Nvidia graphics card or run setup-cpu.py if you do not. NOTE: This splitter will work on a CPU, albeit, very slowly. The reason I keep this option is for people who may want to curate a dataset locally, but train on colab. (AMD not compatible, Mac is not coded for (should be able to use MPS though). Both can use CPU option)
python setup-cuda.py
- Activate the virtual envionrment (venv).
venv\Scripts\activate
-
If you ran into any permission issues, you'll need to change your windows Execution Policy to Remote Signed. This does lower security on your system a small bit as it allows for scripts to be ran on your computer, however, only those signed by a Trusted Publisher or verified by you can be run (to my knowledge). Do at your own risk.
- Open a powershell window as admin. Then, run the following command:
Set-ExecutionPolicy RemoteSigned- If you want to change it back, you can with:
Set-ExecutionPolicy Restricted -
Now rerun step 5 and activate your venv. After it's activated, you can then run the following command to start up the script:
python split_audio.py
For more details, please refer to the youtube video.