Co-authored-by: N0x45f <wangzhen45@baidu.com>

* fix dygraph to static for tacotron2, test=doc

* Fix dy2st error for taco2

* Update attentions.py

---------
Co-authored-by: N0x45f <wangzhen45@baidu.com>

* remove fluid api and useless import.

* fix variable name

* add simple adeadelta optimizer.

* remove useless log

* remove useless and fluid import.

* add framework.dygraph_only back

* rm transformers import and modify variable name consistent with infer.py

* add condition ctc_prefix_beam_search decode.

* Update run.sh

* Update README.md

Co-authored-by: TianYuan <white-sky@qq.com>

* add diffusion module for training diffsinger

* add wavenet denoiser final conv initializer

* add diffusion noise clip to optimize sample result

* clean dependencies

* update paddleaudio's version

* rm dependency in librosa and paddlenlp

* rm dependency in paddlepaddle

* rm dependency in speech_web

* Avoid using variable "attn_loss" before assignment

* Update tacotron2_updater.py

---------
Co-authored-by: TianYuan <white-sky@qq.com>

* add diffusion module for training diffsinger

* add wavenet denoiser final conv initializer