How to apply for PopCS dataset about diffsinger HOT 3 CLOSED

The dataset & pre-trained model for SVS will be made public before the ending of AAAI-22 conference.
We are now working on preprocessing PopCS to obtain more accurate duration labels, segmentations of each song, and F0 labels.
DM001 contains 50 songs, which is only half of PopCS. But maybe you can try DM001, since it seems to have fine annotation.

from diffsinger.

The dataset & pre-trained model for SVS will be made public before the ending of AAAI-22 conference.
We are now working on preprocessing PopCS to obtain more accurate duration labels, segmentations of each song, and F0 labels.
DM001 contains 50 songs, which is only half of PopCS. But maybe you can try DM001, since it seems to have fine annotation.

Is the data format of training DiffSinger as same as training TTS model, such as librispeech mentioned in the readme？

from diffsinger.

The dataset & pre-trained model for SVS will be made public before the ending of AAAI-22 conference.
We are now working on preprocessing PopCS to obtain more accurate duration labels, segmentations of each song, and F0 labels.
DM001 contains 50 songs, which is only half of PopCS. But maybe you can try DM001, since it seems to have fine annotation.

Is the data format of training DiffSinger as same as training TTS model, such as librispeech mentioned in the readme？

You can adjust the logic for data packing by yourself to fit your new singing dataset, or wait for our codes with data.

from diffsinger.