KeyError: 'position_dependent_phones' #18
Replies: 7 comments 19 replies
-
This would be a bug in MFA not with the dictionary. Can you run with --verbose and --debug and paste the full error message to see where it's happening? I'm trying wrap up MFA 2.1 which is the version the japanese model was trained, so this should get resolved once I can release 2.1. |
Beta Was this translation helpful? Give feedback.
-
Ok I think I tracked down where it was throwing the error, and I've updated the zip file for release. Can you redownload the acoustic model via mfa or downloading from https://github.com/MontrealCorpusTools/mfa-models/releases/tag/acoustic-japanese_mfa-v2.0.1a and see if that fixed it? |
Beta Was this translation helpful? Give feedback.
-
It seems now that I have installed the upgraded MFA (2.1), but it starts spitting another kind of error message. Now the error persists for the English annotation files that worked well in the previous version. This may not be to do with the issue of the Japanese model we're talking about, but would you mind looking into the error message if possible? The corpus structure should not be a problem - it has annotated textgrids for each wav file, the file names corresponding to each other. I haven't changed anything from running the aligner in the previous version. It used to both validate and align the folder ok. Thanks so much in advance for your help! (aligner) takayukinagamine@TakayukinoMacBook-Pro ~ % mfa validate /Volumes/Samsung_T5/PhD_Data_Analysis/data/8_forced_alignment/believe/corpus english_us_arpa english_us_arpa |
Beta Was this translation helpful? Give feedback.
-
Can you try updating to the latest 2.1.2? I did add an extra check so that
it wouldn't try to write the file if the directly doesn't exist, and fixed
a few other bugs.
…On Tue, Feb 7, 2023, 5:22 AM Takayuki Nagamine ***@***.***> wrote:
If this helps, MFA keeps deleting the directory
/Users/takayukinagamine/Documents/MFA/corpus/ whenever I create the
folder manually, returning the 'No such file or directory' error above. And
note that I'm installing/trying to run the aligner on the local hard drive,
having disconnected it from iCloud now.
—
Reply to this email directly, view it on GitHub
<#18 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAVJOT6MSJQGR6MVA7EMKH3WWJD75ANCNFSM6AAAAAAUMTABRI>
.
You are receiving this because you commented.Message ID:
<MontrealCorpusTools/mfa-models/repo-discussions/18/comments/4893507@
github.com>
|
Beta Was this translation helpful? Give feedback.
-
I've tried |
Beta Was this translation helpful? Give feedback.
-
Hello Michael, Sorry that it's been a while since the last post but been a little busy getting something else sorted from February until last week. Good news, MFA is working for my Japanese data! Thank you so much for all this. There are some glitches but they may be to do with the wav files. Some of the words I'm forced-aligning do not seem to be included in the dictionary so I may have to do them separately. Will try this a little more and let you know if any issues persist. |
Beta Was this translation helpful? Give feedback.
-
Sounds good, glad to hear it! I am planning on rolling out some better Japanese (and other languages) support with doing tokenization via spacy's https://spacy.io/usage/models. There are a couple of bigger prerequisite changes that have to happen before then, but ideally you should just be able to use regular Japanese text as input and it'll generate tokenized transcripts. |
Beta Was this translation helpful? Give feedback.
-
Hello everyone, I would be grateful if I could seek any help regarding the error "KeyError: 'position_dependent_phones'" when aligning Japanese data. I'm trying to forced-align audio files, each of which contains a single word in Japanese, but I keep getting the error message KeyError: 'position_dependent_phones'. I thought it might be to do with the dictionary, as the pre-trained dictionary does not contain the words in my corpus, but developing a new dictionary didn't help. Does anyone possibly know what may possibly be causing this? I'm not sure what levels of details I need to give but I'd be happy to provide further information if helpful.
Version: MFA 2.0.6
Language/acoustic model: v2_0_1a
Dictionary: Japanese MFA dictionary v2_0_1a then I trained my own based on the corpus using Japanese MFA G2P model v2.0.1a
Thanks,
Takayuki Nagamine
Beta Was this translation helpful? Give feedback.
All reactions