This repository is dedicated to the implementation of specialized neural networks for voice conversion in the context of a master's thesis. Each directory within this repository contains code and resources tailored for training these neural networks. To effectively train the models, it is essential to provide a training dataset with the following specific structure. Please note that the models exclusively accept audio files in the WAV format.
- speaker1
- --sample1.wav
- --sample2.wav
- ...
- speaker2
- --sample1.wav
- --sample2.wav
- ...
- ...
- Cyclegan-basic - contains code for early tests for CycleGAN networks for voice conversion
- INCEPTION-VC-VAE-2 - source code for INCEPTION-VC2 model training and inference
- INCEPTION-VC-VAE - contains code for basic INCEPTION-VC model training and inference
- Transformer-VC-Phase - source code for Transformer-VC training and inference with phase processing
- Transformer-VC-modified - contains code for Transformer-VC training and inference with reduced/extended channel size
- Transformer-VC - source code for basic Transformer-VC training and inference
- Transformer-VGG-for-PHASE - contains code for Transformer-VC training and inference with phase processing by VGG model
- Variational-Autencoder-VC - source code for basic VAE-VC model training and inference
- app - contains source code for running web application for voice conversion based on VAE-VC model
To run test, specific Python environment is required with venv installed.
To create a virtualenv use the following command in specific directory:
python -m venv <environment-name>
To activate created environment in Windows operating system, navigate to <specific-directory>/<environment-name>/Scripts and type Activate.bat.
Once environment run, navigate toVoice-Conversion-Tests repository and type pip install -r requirements.txt.
To run model training, navigate to model directory from Directory contents and run python main.py. Change of config.json file may be required.
Once training is finished, to run inference on model run python generate_sound.py. File generate_sound.py change may be also required to set specific paths.