* Add requirements.txt and pinyin.txt needed by zipvoice * simplify the requirements for pretrained model inference