/Gezi-Opera-Synthesis

Primary LanguagePython

Gezi Opera Synthesis

The official implementation of our AAAI24 paper: FT-GAN: Fine-grained Tune Modeling for Chinese Opera Synthesis.

🗃️ Environments and Dataset

  1. Create an environment with anaconda for example:
conda create -n gezi_opera python=3.8
conda activate gezi_opera
git clone https://github.com/double-blind-pseudo-user/Gezi_opera_synthesis
cd Gezi_opera_synthesis
pip install -r requirements.txt
  1. Download the pretrained vocoder and the pitch extractor, unzip these two files into checkpoints before training your acoustic model.
  2. Download the dataset and unzip it into data/processed.

📄 Preprocessing

Run the following scripts to binarize the data:

export PYTHONPATH=.
CUDA_VISIBLE_DEVICES=0 nohup python data_gen/tts/bin/binarize.py \
--config usr/configs/gezixi.yaml \
> data_processing.log 2>&1 &

The binarized data will be saved to data/binary.

✏️ Training

Run the following scripts to train the model:

CUDA_VISIBLE_DEVICES=0 nohup python tasks/run.py \
--config usr/configs/gezixi.yaml --exp_name your_experiments_name --reset \
> training.log 2>&1 &

🔊 Inference

When training is done, run the following scripts to generate audio:

CUDA_VISIBLE_DEVICES=0 nohup python tasks/run.py \
--config usr/configs/gezixi.yaml --exp_name your_experiment_name --reset --infer \
> inference.log 2>&1 &

Inference results will be saved in checkpoints/your_experiment_name/generated_ by default.