search

openvpi / DiffSinger

An advanced singing voice synthesis system with high fidelity, expressiveness, controllability and flexibility based on DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism
Apache License 2.0
2.73k stars 288 forks source link

关于唱法模型数据集 #194

Open lhc991025 opened 5 months ago

lhc991025 commented 5 months ago
image

要训练唱法模型，可以用公开的opencpop数据集吗？但是它里面的transcriptions.txt文件和你所描述的transcriptions.csv文件不太符合，所以是不能直接用opencpop数据集进行训练吗？

yqzhishen commented 5 months ago

opencpop的原始标注格式有问题，所以要用也得用转换过的。

但依然不建议用它训练，因为它的标注质量不咋地

contiunity commented 2 months ago

不建议用opencpop本身的标注训练，要用opencpop建议自己重新标注