Fix utils/split_scp.pl breakage

[kkm000]

Close #3306

[kkm000]

Hold on, did not cleanly merge

[kkm000]

This is ready to go now. I also cleaned up error messages, and fixed the -j M N argument check with and without one-based.

[danpovey]

Thanks! Note: this is a fixed version of #3285 which was reverted in #3307.

[kkm000]

Yes, I noticed, I clashed with the revert :)

[danpovey]

You sure, right?

[kkm000]

I just used this in a practical script, and updated my scripts to --one-based between two runs, both went out ok; I have thorough error checks in my scripts to make sure data counts all match after processing, and they do. I do not have a practical utt2spk-based case to test at the moment, though (although the next step, merging alignments and lattices, that we once talked on the list, likely will use it).

[kkm000]

You sure, right?

So pretty much yes, and I am likely to use it more a few times today. I'll be merging a complex dataset, with lattices coming from separate alignments, so if anything comes up, I'll notice.

[danpovey]

Remind me to-morrow, I'll merge then.

…

On Thu, May 9, 2019 at 2:03 PM kkm (aka Kirill Katsnelson) < ***@***.***> wrote: You sure, right? So pretty much yes, and I am likely to use it more a few times today. I'll be merging a complex dataset, with lattices coming from separate alignments, so if anything comes up, I'll notice. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#3308 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAZFLO6HBMQNIIG5NAW5TILPURRQDANCNFSM4HL4OLRQ> .

[kkm000]

@danpovey:

Remind me to-morrow, I'll merge then.

Yup. You did the right thing. :) There is more weirdness, fixing now. I'll push when all modes work, and then let you know.

[kkm000]

@danpovey, this flies! (and with the updated combine_ali_dirs, which comes tomorrow).

[danpovey]

OK. @kkm000, FYI I usually prefer to use filter_scp.pl using the appropriate utt2spk file, instead of split_scp.pl. It probably doesn't make much difference though.

[kkm000]

@dpovey, so can we merge it?