Voice Conversion

RPipe

Research Pipeline

Requirements

See requirements.txt

Instructions

• Use make.sh to generate run script
• Use make.py to generate exp script
• Use process.py to process exp results
• Hyperparameters can be found in config.yml and process_control() in module/hyper.py

Examples

• Generate run script

  bash make.sh

• Generate run script

  python make.py --mode base

• Train with MNIST and linear model

  python train_model.py --control_name MNIST_linear

• Test with CIFAR10 and resnet18 model

  python test_model.py --control_name CIFAR10_resnet18

• Process exp results

  python process.py

Examples for VCTKTime

• Generate run script

  bash make.sh

• Generate run script

  python make.py --mode base

• (Optional: if you already have VCTK downloaded) Skip download step

  mkdir data/VCTK
  mkdir data/VCTK/raw

  add 'wav48_silence_trimmed' and 'speaker-info.txt' into the 'data/raw' directory

• Train with VCTK and linear model

  python train_model.py --control_name VCTKTime_linear

• Test with CIFAR10 and resnet18 model

  python test_model.py --control_name CIFAR10_resnet18

• Process exp results

  python process.py

Results

• Learning curves of MNIST

• Learning curves of CIFAR10

Acknowledgements

Enmao Diao