Test on chinese videos

[QUTGXX]

I tried my own videos based on the Colab. The result seems not correct. Do you have any suggestions if I want to train it based on my own datasets?

[v-iashin]

If you are trying to synchronise a talking head in Chinese, then it might not work due to the domain gap (a different language).

I guess you could try to make your dataset resemble the LRS3 (no face crop) structure and train/fine-tune the model.