fighting41love/become-yukarin

Become Yukarin: Convert your voice to favorite voice

Become Yukarin is a repository for voice conversion with a Deep Learning model. By traingin with a large amount of the original and favorite voice, The Deep Learning model can convert the original voice to the favorite voice.

Japanese README

Supported environment

• Linux OS
• Python 3.6

Preparation

# install required libraries
pip install -r requirements.txt

Training

To run a Python script for training, you should set the environment variable PYTHONPATH to find the become_yukarin library. For example, you can execute scripts/extract_acoustic_feature.py with the following command:

PYTHONPATH=`pwd` python scripts/extract_acoustic_feature.py ---

First Stage Model

• Prepare voice data
  • Put input/target voice data in two directories (with same file names)
• Create acoustic feature
  • scripts/extract_acoustic_feature.py
• Train
  • train.py
• Test
  • scripts/voice_conversion_test.py

Second Stage Model

• Prepare voice data
  • Put input/target voice data in two directories
• Create acoustic feature
  • scripts/extract_spectrogram_pair.py
• Train
  • train_sr.py
• Test
  • scripts/super_resolution_test.py
• Convert other voice data
  • Use SuperResolution class and AcousticConverter class
  • sample code

Reference

• ipynb branch: Other sample code
• Commentary Blog (Japanese)
• Realtime Yukarin: Real-time voice conversion system

License

MIT License