search

MontrealCorpusTools / mfa-models

Collection of pretrained models for the Montreal Forced Aligner
Creative Commons Attribution 4.0 International
103 stars 19 forks source link

'KaldiProcessingError' Error constructing TableReader: #19

Open megseekosh opened 1 year ago

megseekosh commented 1 year ago

Hi,

I've tried just about every solution I can think of (reinstalling MFA, attempting on different machines and OS, updating packages, running different versions of python in the virtual env, etc.) and I continue to get a 'KaldiProcessingError' which manifests in the align.2.1.log file. This happens with some, but not all, language models (currently happening with the pre-trained Greek and when I attempt to train a Cantonese one). I'm certain I'm not the first one with this issue, so there must be an easy fix but the 16-bit suggestion is not the problem and I'm completely out of ideas of things to try.

mmcauliffe commented 1 year ago

Can you attach the align.2.1.log file, the compile_train_graphs.2.1.log file, and the make_mfccs.1.log file in the {corpusname}/split{numjobs}/log directory? Not sure which one might be the culprit.

megseekosh commented 1 year ago

Thank you! Hopefully it's a quick fix.

align.2.1.log compile_train_graphs.2.log make_mfcc.1.log

Note that this particular attempt was performed over just a few files, as I was trying to isolate the source of the error, but it manifests over larger corpora as well.

mmcauliffe commented 1 year ago

Can you send the compile_train_graphs.1.log instead (I misspecified it in my original message, sorry!)? The align.2.1.log is complaining about something going wrong with compiling those, so that should have more insight.

megseekosh commented 1 year ago

Yes, here it is:

compile_train_graphs.1.log

Thank you.

mmcauliffe commented 1 year ago

A couple of other potential solutions, if you can try adding the --debug flag and see if that works? I thought I fixed this issue in one of the recent releases, so upgrading to 2.2.3 if you're not already there might fix it, but it should have been fixed in 2.1.2.