ValueError: math domain error when use Dictionaries with pronunciation probability

[yt605155624]

Hi I use MFA 1.0.1 in THCHS-30 (a Chinese datasets) my lexicon are like follows: my *.lab are like follows: then I run the MFA, and get following error:

Setting up corpus information... Creating dictionary information... Traceback (most recent call last): File "aligner/command_line/train_and_align.py", line 171, in File "aligner/command_line/train_and_align.py", line 67, in align_corpus File "aligner/aligner/base.py", line 85, in init File "aligner/aligner/base.py", line 88, in setup File "aligner/dictionary.py", line 410, in write File "aligner/dictionary.py", line 701, in _write_fst_text ValueError: math domain error [22656] Failed to execute script train_and_align

could you please tell me how to solve this?

[yt605155624]

oh ，it's maybe some probability is equal to 0.0

[mmcauliffe]

Oh interesting, yeah I don't think I have any explicit correction for that, so yeah I'll add some dithering to it before it takes the log.

[nonusx]

The same problem, anyone solved?

[nonusx]

I'm trying to validate russian corpus, and caught the same error.