wenet-e2e/wetts

Whether is this a bug?

Opened this issue · 0 comments

Should a line be add after

prosodys = prosody_prob.argmax(axis=1).tolist()

by:
#prosodys = prosodys[1:-1]

e.g. --text 今天天气很不错哟
original: 今天天气很不错哟 => sil j in1 #0 t ian1 #0 t ian1 #2 q i4 #0 h en3 #2 b u4 #1 c uo4 #0 ^ io5 #4
after fix: 今天天气很不错哟 => sil j in1 #0 t ian1 #2 t ian1 #0 q i4 #2 h en3 #1 b u4 #0 c uo4 #0 ^ io5 #4

bug from:
in L63 other two tokens for BERT were added and are not considerted after
#tokens = ['[CLS]'] + [str(x) for x in text] + ['[SEP]']