PronunciationAcousticMismatchError

[mor...@lalilo.co]

Hi, 

I am trying to use the MFA for the first time, I took the English dictionary from your Git (https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner/blob/master/eng.dict) and the English acoustic model from the last release (https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner/releases). My corpus is a wav file with the related TextGrid.

I got the following error:

bin/mfa_align -e -n ./../corpus/ ./../dict/english_dict.dict ./pretrained_models/english.zip ~/Documents/mfa_test

Setting up corpus information...

Number of speakers in corpus: 1, average number of utterances per speaker: 2.0

Traceback (most recent call last):

  File "aligner/command_line/align.py", line 186, in <module>

  File "aligner/command_line/align.py", line 146, in validate_args

  File "aligner/command_line/align.py", line 88, in align_corpus

  File "aligner/models.py", line 129, in validate

aligner.exceptions.PronunciationAcousticMismatchError: There were phones in the dictionary that do not have acoustic models: 

Failed to execute script align

I thought the dictionary could handle more phones than the acoustic model?

I guess the dictionary and the acoustic model are not aligned but I could not find a dictionary from the release 1.0.0.

Thank you!

Morgane

[Michael McAuliffe]

Hi Morgane,

Sorry about that, the dictionary that was in the Git root (removed now) used an unstressed version of Arpabet, while the acoustic models were trained with a stressed version. You can download a stressed Arpabet dictionary derived from Librispeech from http://mlmlab.org/mfa/dictionaries/english.dict.  I've updated the documentation in a couple of places to point there as well.

Michael

[Morgane Daniel]

Hi Michael,

Thank you for your quick answer, I have changed the dictionnary and I am facing a new error: 

bin/mfa_align ./../corpus/ ./../dict/english_dict.dict ./pretrained_models/e

nglish.zip ./mfa_test

Setting up corpus information...

Number of speakers in corpus: 1, average number of utterances per speaker: 2.0

Creating dictionary information...

Setting up training data...

Calculating MFCCs...

Traceback (most recent call last):

  File "aligner/command_line/align.py", line 186, in <module>

  File "aligner/command_line/align.py", line 146, in validate_args

  File "aligner/command_line/align.py", line 93, in align_corpus

  File "aligner/aligner/pretrained.py", line 71, in __init__

  File "aligner/aligner/pretrained.py", line 117, in setup

  File "aligner/aligner/base.py", line 80, in setup

  File "aligner/corpus.py", line 970, in initialize_corpus

  File "aligner/corpus.py", line 848, in create_mfccs

  File "aligner/corpus.py", line 859, in _combine_feats

FileNotFoundError: [Errno 2] No such file or directory: '/home/morgane/Documents/MFA/corpus/train/mfcc/raw_mfcc.0.scp'

Failed to execute script align

I checked in the folder, in make_mfcc.0.log there is:

copy-feats: error while loading shared libraries: libcblas.so.3: cannot open shared object file: No such file or directory

compute-mfcc-feats: error while loading shared libraries: libcblas.so.3: cannot open shared object file: No such file or directory

extract-segments: error while loading shared libraries: libcblas.so.3: cannot open shared object file: No such file or directory

[Michael McAuliffe]

Hi Morgane,

Sorry it's taken me a bit to get back to this.  If you haven't gotten it working in the meantime, the error should be fixed by running "sudo apt-get install libatlas3-base" (assuming you're on Ubuntu).  If you're on another linux distro, you should be able to find a comparable atlas library in their package repositories.

Cheers,

Michael

[Morgane Daniel]

Hi Michael,

Thank you very much for your time, it works well now!

My next step is to run the aligner with a French dictionary and French pretrained models and then to train my own models.

I will keep you update with my tests.

Cheers,

Morgane