This is a system to detect mispronounced words in natural speech and generate the correct pronounciations in the user's voice.
To replicate the training procedure, first export all visible devices enabled by CUDA to -1. Then change directory to the wavegan folder and run
python3 train_wavegan.py train ./train --data_dir ./data/Apple
For observing previews at defined checkpoints in the /train directory created, run
python3 train_wavegan.py preview ./train --data_dir ./data/Apple
This will take the latest checkpoint available and generate the audio for that.
This is the final report on the project. You may find this saved as Final_Paper___GANs.pdf on the same repository.
This browser does not support PDFs. Please navigate to the PDF on the repo to view it: View PDF.