Closed emilyahn closed 1 year ago
same error
So everything should have worked and it should have exported everything successfully, the DatabaseError at the end is because maintenance workers were running when MFA shut down the postgresql server. I thought I had swallowed the error for successful runs in 2.2.11, but I guess not so I'll take another look. Again, everything with the training looks like it completed fine and ldc_pilot.zip should be exported correctly.
Hi Michael, thanks so much for your input! I see that I am able to get the model file ldc_pilot.zip
out correctly, and mfa align ...
works perfectly. In the meantime, I will ignore the DatabaseErrors.
I'd like to point out though that the information on the train command on this page is misleading, because when I followed this option (mfa train ~/mfa_data/my_corpus ~/mfa_data/my_dictionary.txt ~/mfa_data/new_acoustic_model.zip ~/mfa_data/my_corpus_aligned # Export both trained model and alignments
), the alignments did not get deposited into the output folder. It only worked when I used this command: mfa train --output_directory <output_directory> CORPUS_DIRECTORY DICTIONARY_PATH OUTPUT_MODEL_PATH
.
Lastly, is it correct that running the mfa align
command produces a alignment_analysis.csv
in the output folder while mfa train --output_directory <output_directory> ...
does not produce an alignment analysis file?
Oh yes, the syntax for exporting the alignments changed with the updated CLI a bit ago, I'll fix up those docs, and add the alignment analysis export for train.
same error
Debugging checklist
[ yes; 2.11] Have you updated to latest MFA version? [ yes] Have you tried rerunning the command with the
--clean
flag?Describe the issue Upon training an acoustic model (on 38 min of English TIMIT) from scratch, there is an error about the database server. This issue also occurs if I try to run
mfa validate
.For Reproducing your issue Please fill out the following:
Log file pg_log_global.txt terminal output ends like this:
Desktop (please complete the following information):
Additional context I'm also doing a thing where each word in the lexicon and text input is a phone, since TIMIT has manually aligned phone labels. I don't know if this is throwing things off. For example, here is a .lab file:
sil p iy t s er r iy ih z aa r k ax n v iy n ih t f er k w ih k l ax n ch sil
And this is a snippet of the short lexicon file: