Jiachen Lian

To complement,   As you can see, when loss goes negative, loss still decreases. But results seem to be better?

If you put only one wav file into each speaker directory, this modified make_metadata.py should work: ``` import pickle from model_bl import D_VECTOR from collections import OrderedDict import numpy as...

I think the thing is there is no overlap between training set and test set. Labels of training set vary from 0 to 8612 and 0 to 499 for test...

I think there may be some problems in this: command = ("ffmpeg -y -i %s -qscale:v 2 -async 1 -r 25 %s" % (opt.videofile,os.path.join(opt.avi_dir,opt.reference,'video.avi'))) output = subprocess.call(command, shell=True, stdout=None)

> Yeah, the `rdzv_id` was the cause for that error, which should be the same for all nodes, I should've read the docs more carefully. But I think this line...

你好，我并不是这个paper的作者，不过我可以找找看有没有数据