Implements FastSpeech1/2 models via PyTorch. Wandb report
- hw_tts/ is main module of this project; it directly implements FastSpeech and FastSpeech2.
- requirements.txt lists all python requirements.
- preprocess_data.py implements data preprocessing.
- setup.sh script installs needed requirements, downloads data and preprocesses it.
- train.py and synthesis.py scripts are used to train model and synthesise audio respectively.
- launcher.ipynb notebook shows possible ways to run scripts for this project.
- figs/ directory contains some figures for report.
To install all requirements, download needed data and preprocess it simply run:
./setup.sh
To train model run
python3 ./train.py -c ./hw_tts/configs/fs2_train_config.json
To synthesise audio run
python3 ./synthesis.py -c ./hw_tts/configs/fs2_synthesis_config.json -w ./fs2_saved/checkpoint_19999.pth.tar
All scripts menthioned above are also shown in launcher.ipynb notebook.