Implementation of the paper One-shot Voice Conversion by Separating Speaker and Content Representations with Instance Normalization modified from the official one and heavily based on https://github.com/cyhuang-tw/AdaIN-VC.
Requires at least python 3.6. For other dependencies, see requirements.txt.
The main difference from the official implementation is the improvements of audio quality due to the use of a neural vocoder (universal vocoder, whose code was from yistLin/universal-vocoder).
Besides, this implementation supports torch.jit, so the full model can be loaded with simply one line:
model = torch.jit.load(model_path)The main contribution of this repo is organizing instructions for preprocessing, training and inference into a runnable notebook.
Also, usage instructions are automatically available using python adain-vc.py --help thanks to python-fire.
Please cite the paper if you find AdaIN-VC useful.
@article{chou2019one,
title={One-shot voice conversion by separating speaker and content representations with instance normalization},
author={Chou, Ju-chieh and Yeh, Cheng-chieh and Lee, Hung-yi},
journal={arXiv preprint arXiv:1904.05742},
year={2019}
}