Unusual behavior in Subtitle Edit Whisper voice to music transitions

[loninappleton]

This is another hard to explain behavior. I'm wondering when the transitions in and out of a job where Whisper prints [music] over dialog will be available.

[music]
[music]
[music]

is all that prints after _some_ sections of speech and dialog. It just goes to [music] and never comes back out of it when dialog resumes. That could go to the end of the piece for a half hour or more. It's not consistent that I can see.

I'm going to see if I can get dialog from VOSK for the missing non printing dialog parts and possibly hand copy.

[Subtitles]

VOSK will not have this kind of behaviour.
I mentioned in another post that you imitiated that Whisper AI can get into endless loop and in the help menus there is a solution when using the command line version of Whisper AI.
Unfortunately it seems that SubtitlesEdit sticks to basic default settings and doesn't have the flexibility that the command line version has.
Perhaps in newer versions the developer might add them to make SE better. You can contact him and ask him if this is possible.
Anyway, this is the option that has helped me to get the transcription out of endless loop.
--condition_on_previous_text CONDITION_ON_PREVIOUS_TEXT
if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop (default: True)

This is what add to my command line:

--condition_on_previous_text False

[Subtitles]

I did some digging today and if you want to to be able to get Whisper not going crazy with repeated text over and over, then you might want to try this variation for Whisper.
It is called whisper-faster r128 and works in command line without the need to install Python.
https://github.com/Purfview/whisper-standalone-win/releases
Download the zip file.
You will need to install few files to be able to use a GPU, even if you don't have one
https://github.com/Purfview/whisper-standalone-win/releases/tag/libs
Download the zip file.
Extract the two zip files to c:\ drive and have the GPU files in the same directory as whisper-faster.exe
I am attaching here the help file for whisper-faster.
Put a short test audio file in the directory and try to transcribe it.
You should add --device cpu and --condition_on_previous_text False in the command line.
It will take a little while so check the directory if it creates an srt file, as it doesn't tell you when it is finished.
Let me know if this helps.

[loninappleton]

Yes I know I mentioned it before... just wondering about any progress.
The best advice for me is to contact Nikse and hope for some changes.

[Subtitles]

Yes I know I mentioned it before... just wondering about any progress.
The best advice for me is to contact Nikse and hope for some changes.

I doubt it. You are probably using the cpp option which is a 3rd party developed software, and he is just incorporating this into his GUI.
He will tell you to use Whisper AI or any of the other options besides cpp.
cpp doesn't have many options for transcribing. It the the simplest one.
If you look at the other options besides cpp, there are few more all needing Python and a good GPU.
If you have the time, try to use whisper-faster.

[loninappleton]

Indeed cpp is on the upper right hand corner-- I guess by default.
I'll avoid asking direct then... but at SE there is a team of contributors referenced with each release. I would think someone would address the issue amidst all the fixes.

It's not the only question in using SE for this task. I don't know what the check box to fix errors actually does. But I've seen in VOSK that combining lines via fix errors is mandatory for any sensible output.

Regardless I have a job going that'll take some time-- handwork corrections. If anyone thinks that Whisper is the last thing to do with subs, I've got a little list for them.

[VoodooFX]

It will take a little while so check the directory if it creates an srt file, as it doesn't tell you when it is finished.

It writes when it's finished, even gives a progress bar with ETA.

Maybe I'll add some sound alarm in next release.

[loninappleton]

Perhaps someone will test this procedure and report back any success.

[Subtitles]

It will take a little while so check the directory if it creates an srt file, as it doesn't tell you when it is finished.

It writes when it's finished, even gives a progress bar with ETA.

Maybe I'll add some sound alarm in next release.

First test it didn't show when it finished but then it did.
Yes a sound alarm would be nice to have.
Great software thank you for developing it.