So, if I understand, you have unicode characters in the utt2spk?
Thanks for the fix!
y.

On Fri, Oct 28, 2016 at 4:08 AM, 薛丞宏 notifications@github.com wrote:

utils/scoring/wer_per_spk_details.pl deals with STDIN, STDOUT and $ARGV[0],
three I/Os.
The origin script set STDIN ad STDOUT only.

binmode STDIN, ":utf8";
binmode STDOUT, ":utf8";

So $ARGV[0] should be set.
The code use open qw(:std :encoding(UTF-8)); was copied from stackoverflow
http://stackoverflow.com/a/14566813/3640653.

PS: I found this problem because I got the warnings below:

Use of uninitialized value $SPK in hash element at utils/scoring/wer_per_spk_details.pl line 118, line 4.
Use of uninitialized value $SPK in hash element at utils/scoring/wer_per_spk_details.pl line 119, line 4.
Use of uninitialized value $SPK in hash element at utils/scoring/wer_per_spk_details.pl line 120, line 4.
Use of uninitialized value $SPK in hash element at utils/scoring/wer_per_spk_details.pl line 121, line 4.
Use of uninitialized value $SPK in hash element at utils/scoring/wer_per_spk_details.pl line 122, line 4.
...

---

You can view, comment on, or merge this pull request online at:

#1149
Commit Summary

• wer_per_spk_details.pl support that utt2spk contains unicode

File Changes

• M egs/wsj/s5/utils/scoring/wer_per_spk_details.pl
  https://github.com/kaldi-asr/kaldi/pull/1149/files#diff-0 (1)

Patch Links:

• https://github.com/kaldi-asr/kaldi/pull/1149.patch
• https://github.com/kaldi-asr/kaldi/pull/1149.diff

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1149, or mute the thread
https://github.com/notifications/unsubscribe-auth/AKisXwV4LWOrOH3CiIzN5miwySdn_izCks5q4a17gaJpZM4KjLpY
.

Yes, some speaker names contain unicode chars.

@danpovey I guess this is fine to commit -- I'd never assumed that spk labels would contain unicode.