torch-seq2seq-singing-synthesis

pytorch implementation of https://mtg.github.io/singing-synthesis-demos/transformer/

dataset used：https://wenet.org.cn/opencpop/

to try:

• edit ROOT_PATH in config.py
• create a folder named "raw" in ROOT_PATH and drop wavs and textgrids in pairs
• run preprocess/cut_wav.py
• run prepare_train_data.py
• run trainer.py

todo：

• [ ] tensorboard
• [ ] hifi-gan vocoder
• [ ] pitch model