Parallel waveform generation with DiffusionGAN

Parallel waveform generation with DiffusionGAN

• DiffWave: A Versatile Diffusion Model for Audio Synthesis, Zhifeng Kong et al., 2020. [arXiv:2009.09761]
• Tackling the Generative Learning Trilemma with Denoising Diffusion GANs, Xiao et al., 2021. [2112.07804]

Requirements

Tested in python 3.7.9 conda environment.

Usage

Download LJSpeech dataset from official:keithito.

To train model, run train.py

python -m utils.dump 
    --data-dir /datasets/ljspeech 
    --output-dir /datasets/ljspeech/vocdump 
    --num-proc 8

python train.py 
    --data-dir /datasets/ljspeech/vocdump 
    --from-dump

GitHub

View Github