FastSpeech project barebones

Installation guide

Copy this repo.

pip install -r ./requirements.txt
gdown https://drive.google.com/uc?id=1SzRWUMCG6LaiyM9VYtmBe4TW5itennWP -O final_data/model.pth
bash gen_scripts/downloader
python3 gen_scripts/energy_generation.py
python3 gen_scripts/pitch_generation.py

Model is located in directory hw_sg/model/gen_model and consists out of three blocks. transformer.py for transformer blocks, predictors.py for variance adapter and FastSpeech.py for final model.

We also modified trainer.py, ljspeech_dataset.py and collate.py in order to make our model work.

Also we have some generation scripts in directory gen_scripts and we shall run them on startup.

final_data/config.json is a trainer and tester config.

In order to test our model you should call

python3 test.py -c final_data/config.json -r final_data/model.pth

Needed texts and energies can be changed manually in test.py. Results will be saved to results folder.

Training is standard with train.py.

Wandb Report and Samples

Link to audio Samples

Link to report