Stuck UI with Long Files

[architeck]

I'm transcribing 2+ hour long wav files and notice that the transcription is completed rather quickly on a Nvidia 3070 using large-v3-turbo model. I can see the completed transcript in the output folder, but the UI remains stuck on "transcribing".

The image below is shows less time, but luckily I checked the folders after my long file run for 5+ hour without and progress.

Edit: I'm actually noticing the stuck UI thing on smaller files as well now.

[roperna]

I've been facing the same issue in the last three days. I'm transcribing 1/1.5 hour long wav files. Compared to last week, I cannot select the model anymore (the only option now is 'large-v3-turbo'), and the UI remains stuck on "transcribing". I checked the folders after my files running for 10+ hours, but there are no progress.

[JuergenFleiss]

Ok, we released the new version 1.2 three days ago, so this is probably a bug that we did not catch in testing and I cannot reproduce it on my end right now. We will have to take a look on the new way we track the progress in the new steps.

Large-v3 turbo is freshly released from OpenAI and offers great balance between speed and accuracy. As for the other models, you now have a new menu point on the left: Model Manager. There you can download all the available whisper models, also large-v2 or large-v3.

Could you try other models and let me know if the UI is also stuck with those?

[SchwarzDeveloping]

Same here. It randomly gets stuck with files over 23 minutes long, while files under 20 minutes pose no issues. The file extension doesn't seem to matter.

[JuergenFleiss]

Same here. It randomly gets stuck with files over 23 minutes long, while files under 20 minutes pose no issues. The file extension doesn't seem to matter.

Ok, I will try to reproduce it.

In the meantime, you can revert back to v1.1 using the installer here: https://business-analytics.uni-graz.at/de/forschung/atrain/download/

[stevevaius2015]

Hello, Did you release v1.2 ? Where is the release to download for developers?

[JuergenFleiss]

Hello, Did you release v1.2 ? Where is the release to download for developers?

Hi, we released the new 1.2 installer on microsoft store last week; I also did a release on github now.

[aereimer]

Hi @JuergenFleiss, thanks for the amazing software. Just want to report that I got similar issues using both atrain 1.2 and atrain core 1.2 (using this command line --model "large-v3" --language "en" --speaker_detection --num_speakers "auto-detect" --device "GPU" --compute_type "float16") in my case, changing the model (turbo, large-v3) did not matter It seems 1.1 is working fine - UPDATE: 1.1 crashed too. im investigating if it is an issue with the mp3 file. It was a 2:13h audio; splitting it worked for some files. some chunks show this error:

Command '['aTrain_core', 'transcribe', 'C:\Users\Data analysis\Downloads\segments\output_003.mp3', '--model', 'large-v3', '--language', 'en', '--device', 'GPU', '--compute_type', 'float16', '--speaker_detection', '--num_speakers', 'auto-detect']' returned non-zero exit status 3221226505.

[JuergenFleiss]

Ok, thats quite strange. Might be related to faster-whisper itself, they also have reports of videos >20min crashing https://github.com/SYSTRAN/faster-whisper/issues/687

Could one of you share the files that crash atrain so that we can investigate (as we are unable to reproduce the problem)? You can reach me under bandas@uni-graz.at