error with generating keyframe

[enigma-dy]

LOG select -r POLYWINK_Louise ; import auto_lip_sync auto_lip_sync.start()

b'Setting up corpus information...\r\n' b'Number of speakers in corpus: 1, average number of utterances per speaker: 1.0\r\n' b'Creating dictionary information...\r\n' b'Setting up training data...\r\n' b'Calculating MFCCs...\r\n' b'Calculating CMVN...\r\n' b'Number of speakers in corpus: 1, average number of utterances per speaker: 1.0\r\n' b'Done with setup.\r\n' b'Done! Everything took 6.46065616607666 seconds\r\n' b'\r 0%| | 0/2 [00:00<?, ?it/s]\r 50%|##### | 1/2 [00:01<00:01, 1.03s/it]\r100%|##########| 2/2 [00:02<00:00, 1.18s/it]\r\n'

Traceback (most recent call last):

File "C:\Users/solomon/Documents/maya/2024/scripts\auto_lip_sync\auto_lip_sync.py", line 322, in generate_animation

self.create_keyframes()

File "C:\Users/solomon/Documents/maya/2024/scripts\auto_lip_sync\auto_lip_sync.py", line 373, in create_keyframes

tg = textgrid.TextGrid.fromFile(textgrid_path)

File "C:\Users/solomon/Documents/maya/2024/scripts\textgrid\textgrid.py", line 807, in fromFile

tg.read(f)

File "C:\Users/solomon/Documents/maya/2024/scripts\textgrid\textgrid.py", line 698, in read

encoding = detectEncoding(f)

File "C:\Users/solomon/Documents/maya/2024/scripts\textgrid\textgrid.py", line 87, in detectEncoding

with codecs.open(f, 'r', encoding='utf-16') as source:

File "C:\Program Files/Autodesk/Maya2024\Python\lib\codecs.py", line 905, in open

file = builtins.open(filename, mode, buffering)

FileNotFoundError: [Errno 2] No such file or directory: ''

[adamgravois]

I am having the same error, currently trying to track it down. I have one wav + txt pair that works, and another that doesn't. Here is what I know: when it fails, it's because MontrealForcedAligner isn't creating a .TextGrid file Why not? IDK!

MFA log messages amount to, "yep, it failed" but never why: tri_fmllr/log/align2.0.log Good run: (gmm-align-compiled[5.4-win]:main():e:\dev\tools\kaldi\src\gmmbin\gmm-align-compiled.cc:135) Overall log-likelihood per frame is -96.5823 over 1166 frames. tri_fmllr/log/align2.0.log Bad run: (gmm-align-compiled[5.4-win]:main():e:\dev\tools\kaldi\src\gmmbin\gmm-align-compiled.cc:135) Overall log-likelihood per frame is -nan(ind) over 0 frames. OK, nan = Not A Number, so that ain't good.

But why? It isn't the audio format. It isnt the text file. Is there something weird about that piece of audio? I tried re-recording with a different voice, and no joy.

If I come up with a solution (upgrade to MFA v3, lol) I'll update this post.

[adamgravois]

update: Running a longer piece of audio (18 sec vs 10 sec) seems to avoid the problem. Same speaker, same words, slightly different recording (but not by much).

[joaen]

Interesting. I'm pretty swamped with work at the moment, but I will take a closer look when I have more time. Like @adamgravois said it seems to be related to MFA failing for some reason.

I'm gonna see if I can repro this on my end, can someone please share a .wav-file that I can use when debugging this?

[adamgravois]

Possibly an easy fix would be trapping that failure and reporting a meaningful error message "MFA couldn't parse that audio, maybe try a silghtly different clip?"

Goodness knows, I tried to make sense of the MFA documentation, but it is super jargon-y. Very very deep in its own expertise subculture.

Thanks for taking my pull req! Audio Samples for MFA.zip

[EnigmaDY]

I'm sorry i really dont know much about coding or how report a bg or error. I just copied what maya output and pated it here

[joaen]

@adamgravois Thanks for the files! I'll try to repro the issue with the files you provided.

[adamgravois]

I'm sorry i really dont know much about coding or how report a bg or error. I just copied what maya output and pated it here

That's a good first step :)