torch_npss

中文(chinese)

implementation of A Neural Parametric Singing Synthesizer: https://arxiv.org/abs/1704.03809

• pretrained models are provided in snapshots/
• generated samples are in data/gen_wav/

dataset：https://drive.google.com/file/d/137dTlTiN7jSadV76sRDRwNJ_ysTz1psA/view?usp=sharing

sample

try it out!

note: test labels are in data/timbre_model/test
pip install -r requirements.txt 
python inference.py

try with your own data

put your own raw and label data in data/raw/
change custom_test in data/preprocess.py to True
run data/preprocess.py
run generate_test('your_file_name') in inference.py 

train your own model

• put your audio and label in data/raw
• run data/preprocess.py
• adjust condition_channel in hparam.py according to your data
• run train_harmonoc.py train_aperoidic.py train_vuv.py

• if you have any questions feel free to leave an issue

A complete version of implementation which including F0 model and vuv model see here: https://github.com/seaniezhao/cnnpss