OpenVino transcription worked and now doesn't anymore!

Hello!
I tried openvino earlier today and it worked perfectly.
A few hours laters, i tried to transcript an other piece of audio and now the transcription is erratic and looks like it mixes all the languages.
I redownloaded the 2 zips and reinstalled it, uninstalled and reinstalled audacity, re-deleting openvino et re-copying it, rebooting my computer each time, tried to delete the “cache.json” file, nothing worked…

it’s super frustrating given it was perfectly working a few hours before, and all i did on my computer during this time was not audacity related!
i have no error message when running openVino.

here is a screenshot of one of my failed attempts:

and here is the console log:

18:32:26: Audacity 3.4.2
18:32:26: Error: Failed to load shared library 'avformat-60.dll' (error 126: Le module spécifié est introuvable.)
18:32:26: Error: Failed to load shared library 'avformat-59.dll' (error 126: Le module spécifié est introuvable.)
18:32:26: Error: Failed to load shared library 'avformat-58.dll' (error 126: Le module spécifié est introuvable.)
18:32:26: Error: Failed to load shared library 'avformat-57.dll' (error 126: Le module spécifié est introuvable.)
18:32:26: Error: Failed to load shared library 'avformat-55.dll' (error 126: Le module spécifié est introuvable.)
18:32:31: sqlite3 message: (1) no such table: project in "SELECT 1 FROM project LIMIT 1;"
18:32:31: Error: Failed to load shared library 'avformat-60.dll' (error 126: Le module spécifié est introuvable.)
18:32:31: Error: Failed to load shared library 'avformat-59.dll' (error 126: Le module spécifié est introuvable.)
18:32:31: Error: Failed to load shared library 'avformat-58.dll' (error 126: Le module spécifié est introuvable.)
18:32:31: Error: Failed to load shared library 'avformat-57.dll' (error 126: Le module spécifié est introuvable.)
18:32:31: Error: Failed to load shared library 'avformat-55.dll' (error 126: Le module spécifié est introuvable.)
18:32:54: File name is C:\Users\Alice\Documents\01_flow morphee\Session test workflow\YogaDoux.wav
18:32:54: Mime type is *
18:32:54: Opening with libsndfile
18:32:54: Open(C:\Users\Alice\Documents\01_flow morphee\Session test workflow\YogaDoux.wav) succeeded
18:32:55: Operation 'Importation de WAV (Microsoft)' took 0,656000 seconds. Poll was called 126 times and took 0,016437 seconds. Yield was called 5 times and took 0,007739 seconds.
18:33:06: Operation 'Pré-traitement' took 0,000000 seconds. Poll was called 0 times and took 0,000000 seconds. Yield was called 0 times and took 0,000000 seconds.
18:33:27: Operation 'OpenVINO Whisper Transcription' took 21,061000 seconds. Poll was called 508 times and took 0,260547 seconds. Yield was called 22 times and took 0,225659 seconds.

any help would be super welcome!!

Hi @batmad,

Can you try this… remove the following folder:

%appdata%\audacity\openvino-model-cache

(You can copy & paste that into windows explorer as path).

And then after removing it, try to run transcription again. If transcription works, then it seems that you’ve hit some bug that some folks have been reporting, but I’ve been unable to reproduce so far.

Anyway, let me know if that is indeed the case…

Thanks,
Ryan

aaaaand it works perfectly again!
thank you so much!
i’ll update you if the bug produces again and if i find any kind of pattern.
The only think i can think about, is that yesterday after trying OpenVino for the first time (when it worked) i used Reaper, and reaper and audacity were open at the same time during a short period.
Otherwise it was just basic office tasks…

thank you again!

aaaand it doesn’t work anymore :stuck_out_tongue:

the pattern is simple: i delete the model cache folder, it works, i do nothing except relaunching openvino, it doesn’t work anymore!

it’s not a big deal anymore since you gave me the “tip” to make it work again,
but i thought you might want the “blob” file inside the model cache:

@batmad,

Thanks for confirming! Yes, this is the issue that I’ve had trouble reproducing. Basically, the first time a feature is used, the model is ‘compiled’ for that device (which can take a little bit of time), but then ‘cached’ as those .blob files into the folder. And so the ‘second’ time, the model is loaded from the ‘cached’ file – and in your (and a few others) case, it seems like something goes wrong prepping the model from cache, and so it produces garbage as output. This is why you need to keep deleting the folder… this prevents the model getting loaded from cache.

Thanks for sharing the .blob, let me see if I can do anything with that. BTW, can you share the details about which CPU model you’re using, along with specific version of Windows? Maybe it can help me find some machine to reproduce on.

Thanks again!
Ryan

i use windows 10 pro, 2004 version (!!??) 19041.928
i got these infos with winver

“windows feature experience pack” 120.2212.551.0

my computer is an old dell laptop lattitude E5540 with an i5-4300 @1.9gHz

thank you again!

Hi, i have same issue and have Windows10 pro. hope that helps . . .

Hi @batmad, @Zicmuse,

Would you mind sharing what language of Windows installation you are using? It may have nothing to do with this, but it seems that alot of users who have reported this are using non-english versions.

Thanks!
Ryan

french! posts must be at least 10 characters so bla bla bla

I’m french Too, so language is french :slight_smile:

This topic was automatically closed after 30 days. New replies are no longer allowed.