An implementation of Tacotron2 based Pytorch
Input: Chinese Pinyin Sequence
Output: Mel-spectrogram
- Prepare training data: text and corresponding mel-spectrogram files.
Referred：https://github.com/Labmem-Zhouyx/audio2mel_preprocessor
- Train the model.
python train.py -o save -l logs
- Inference.
python inference.py -t 'sentences.txt' -c ./save/checkpoint_100000