AUTOVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss

This repository is reproduced code of "AutoVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss"

The paper link is here

Preparation

Pretrained model of WaveNet vocoder is from author's original repogitory

How to use

./scripts/downloadVCTK.sh

python preprocess.py --wav-dir ./VCTK-Corpus/wav48/

python train.py

python3 inference.py --output ./output.wav \
                     --src-wav [path to src wav] \
                     --src-emb [path to src embedding] \
                     --tgt-emb [path to target embedding] \
                     --vocoder [path to vocoder checkpoint] \
                     --autovc [path to autovc checkpoint]

Sample Audio

source

target

converted