minglyubyte/csci566-narrator-group_project

CSCI566-Narrator-Group-Project

This repository is forked from xcmyz's implementation and updates the model by incorporating a neural discriminator to implement an Alpha GAN architecture.

Prepare Dataset and Models

1. Download and unzip LJSpeech dataset.
2. Put LJSpeech dataset under data folder.
3. Unzip alignments.zip.
4. Put Nvidia pretrained waveglow model under waveglow/pretrained_model folder.
5. Download the trained Alpha GAN model and put it under model_alpha_gan folder.
6. Download the trained FastSpeech model and put it under model_fastspeech folder.
7. Run python3 preprocess.py.

Training

Run python3 train.py.

Evaluate Newly Trained Model

1. Run python3 eval.py.

Evaluate Trained Models in Report

1. Change the checkpoint_path in hparams.py to either model_alpha_gan or model_fastspeech.
2. Run python3 eval.py.

Plot Training Curves

Run python3 plot.py

Paper

• FastSpeech Paper
• Alpha GAN Paper