VOICE CONVERSION USING FEATURE SPECIFIC LOSS FUNCTION BASED SELF-ATTENTIVE GENERATIVE ADVERSARIAL NETWORK FLSGAN-VC

Authors

Sandipan Dhar • Padmanabha Banerjee • Nanda Dulal Jana • Swagatam Das

Dataset links and speech samples are provided below:

Datasets

• CMU Arctic
• VCC2018 Non-Parallel Dataset

Generated Samples

CMU Arctic

• Female to Female
• Female to Male
• Male to Female
• Male to Male

VCC2018

• Female to Female
• Female to Male
• Male to Female
• Male to Male

More Generated samples of CMU Arctic and VCC2018 Dataset

Samples of CMU Arctic and VCC 2018

Code

Prerequisites

• Linux, macOS or Windows
• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Downloading Datasets

Download and save in "Data" folder

• CMU Arctic
• VCC2018 Non-Parallel Dataset

Installation

• Clone this repo:

git clone https://github.com/BlueBlaze6335/FLSGAN-VC.git
cd FLSGAN-VC

• Install all the dependencies by

pip install -r requirements.txt

ipython==8.8.0
librosa==0.9.1
matplotlib==3.4.2
numpy==1.19.5
SoundFile==0.10.3.post1
tensorflow==2.6.0
tensorflow_addons==0.17.0
torch==1.12.0
torchaudio==0.12.0
tqdm==4.42.1

Run

• Training

python train.py

• Testing

python inference.py

Acknowledgments

Our code is heavily inspired by MelGAN VC.