Skip to content
This repository has been archived by the owner on Apr 18, 2022. It is now read-only.

Prosody Mark lead to very strange results #26

Open
crystal0913 opened this issue Nov 23, 2018 · 2 comments
Open

Prosody Mark lead to very strange results #26

crystal0913 opened this issue Nov 23, 2018 · 2 comments

Comments

@crystal0913
Copy link

我用你的这个方法mtts.py生成训练的lab,在标贝数据集上,如果去掉文本中的#1、#2、#3、#4这些韵律标签,用merlin训练能得到一个还算ok的合成结果。
但是如果携带这些韵律标签去生成lab,同样的参数下,合成的效果就很差,差到听不得。请问这是什么原因呢?
是否有Prosody Mark对产生lab的区别只有:
韵律词层:/D:d1=d2!d3@d4-d5& 以及 语句层F中的 f4

@Jackiexiao
Copy link
Owner

训练模型时有使用韵律标签吗?

@crystal0913
Copy link
Author

我是用带韵律标签去生成lab,然后用这些lab去训练时长模型和声学模型的,不知道您是不是指这个意思?

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants