X-CCS/Chinese_TTS

tensorflow for TTS (including Chinese English and other languages)

声学模型训练

1. 根据 TextGrid 提取 duration文件：

python examples/mfa_extraction/txt_grid_parser.py --yaml_path examples/fastspeech2_wanmei/conf/fastspeech2wanmei.yaml --dataset_path ./data/align --text_grid_path ./data/TextGrid --output_durations_path ./data/durations

2. Preprocessing

python tts-preprocess.py --rootdir ./data/align --outdir ./dump_wanmei --config preprocess/wanmei_preprocess.yaml --dataset wanmei

3. Normalization

python tts-normalize.py --rootdir ./dump_wanmei --outdir ./dump_wanmei --config preprocess/wanmei_preprocess.yaml --dataset wanmei

4. 根据 duration 文件 筛选出 所有配对文件， 包括 energy wav pitch 等， train目录和 valid目录都要保证一致

5. Train