Subtitle Edit 3.6.10 new version with Whisper option

[loninappleton]

I have seen some install procedures for Whisper on the youtube for it involving a lot of command line work. But the new version of Subtitle Edit 3.6.10 shows an option for it.

With the previous audio to text VOSK model in SE, I had a problem of having the download that Subtitle Edit provides the embedded link for simply hang up.
That is apparently the case with whatever the install option for Whisper is. I needed the proper path in Windows 7 which I eventually got for VOSK. Once that was done I could unpack VOSK model from Alphacephei where it needed to go.

Is anyone else working with Whisper? No youtube tutorials please. I can't follow them.

If SE can link to Whisper through the program, where does it go to do so? Remember mine just hangs up and I need a download source similar to going to Alphacephei and getting the speech model to download instead.

It may be all over my head but I'm always looking for more accurate audio(voice) to text for subtitling.

[Nikse]

Hi Lon,

Whisper can be a bit tricky to get working... first, some fixes regarding Whisper has been made in latest beta, so if you want to use Whisper, please try the beta: https://github.com/SubtitleEdit/subtitleedit/releases/download/3.6.10/SubtitleEditBeta.zip


Whiper (cpp) should be installed to a "SE data folder"/Whisper folder, see more here: https://www.nikse.dk/subtitleedit/help#audio_to_text
Download whisper cpp from here: https://github.com/SubtitleEdit/support-files/tree/master/whisper (you probably need the AVX2 version - depends on your cpu)

It might not work for all... as can be seen here: https://github.com/SubtitleEdit/subtitleedit/issues?q=is%3Aissue+Whisper

Also, if you have trouble downloading via SE in-program downloading using Windows 7, do read this:
https://stackoverflow.com/questions/63352796/error-in-windows-7-the-request-was-aborte...tls-secure-cha

[loninappleton]

Hello again Nikolaj

Thanks for the reply. I will practice the install of Whisper on my backup PC to see how that goes.
I may have to wait for things to smoothe out with it. I did see the Whisper references in the log for SE 3.6.10.
I appreciate the direct links to the Whisper tools. I'm always appreciative of your help to users.

Lon

[loninappleton]

I have made a Subtitle Edit Beta folder, hopefully getting all the Whisper cpp necessities inside that install which is SE 1.3.6.45 if I make that out right.
The SE 'data folder has to be just the main where dll's are found since there is no folder specifically marked data. I put a copy of ffmpeg in this folder as well. That was another instance where the downloader inside SE hangs. I have to read that link you gave to Substack yet.
Reviewing the install from VOSK, I have the location: C:\users\lon\AppData\Roaming\Subtitle Edit\VOSK\VOSK model

The Whisper ccp should be something similar such as unzip whisper.cpp-AVX2-2022-11-16.zip to C:\users\lon\AppData\Roaming\Subtitle Edit.

The other thing I'm getting is the model itself which is 1.5 Gb called ggml-medium.en.bin which is probably in error. please advise.

I'll come back when I try the unzip to AppData etc etc. have to quit at the moment.

[loninappleton]

My unzip to Subtitle Edit in AppData is not working well. I have to direct it to Subtitle Edit Beta but I don't see it in the AppData Roaming tree.
here is the screen shot of users/lon:

[loninappleton]

Just a note on the stackoverflow piece. It wants me to create a registry key. I'm wary of doing anything like that except on a scratch disk.

[Nikse]

"main.exe"and "whisper.dll" needs to be in a "Whisper" folder.
And you should unpack the beta version to something like "C:\Tools\SE" - and run "C:\Tools\SE\SubtitleEdit.exe"

[VoodooFX]

I had a problem of having the download that Subtitle Edit provides... ...Windows 7...

There I made the quick fix: https://pastebin.com/h8NnpysH

Press a download button there and save with extension "reg" instead of "txt", then double click a reg file.

[loninappleton]

Thanks for all the help on this. It may take me a while to get it right. Since SE 3.6.7.45 is not showing up in
Win7 programs, I'm going to start over with a fresh install so that the AppData unzip goes to the right place.
I did the install from inside a new folder to keep it separate and that must have created an error.
I'll come back in when I've got a fresh SE Beta showing up.

[loninappleton]

Thanks for all the help on this. It may take me a while to get it right. Since SE 3.6.7.45 is not showing up in
Win7 programs, I'm going to start over with a fresh install so that the AppData unzip goes to the right place.
I did the install from inside a new folder to keep it separate and that must have created an error.
I'll come back in when I've got a fresh SE Beta showing up.

Here I'm correcting that it should be and I do have I do have SE 3.6.10

[loninappleton]

I have made a screen shot of what I have so far. I shows Subtitle Edit 3.6.10 NEXT sub 45 in the Win7 Programs folder.

I created a folder called Whisper

I unzipped the whisper-cpp-AVX2-2022-11-16.zip to the whisper folder.

If I seem to be going slow, that's just my way of understanding the steps.

If ok so far, I'll look to procede with the regedt procedure (at some point) and getting the correct whisper speech model. the speech model
I have downloaded is ggml-medium.en.bin.zip. Please give the correct speech model download and a direct link of possible. I'll review the rest of the thread as well.

[loninappleton]

I had a problem of having the download that Subtitle Edit provides... ...Windows 7...

There I made the quick fix: https://pastebin.com/h8NnpysH

Press a download button there and save with extension "reg" instead of "txt", then double click a reg file.

Thanks, found it but made no changes yet. Once the txt is changed to reg, then will it install itself in the Registry? Or will I see the 'gear' symbol or what?


This would be of great advantage to be able to do these SE installs from the program as Nikse has provided.

[VoodooFX]

Thanks, found it but made no changes yet. Once the txt is changed to reg, then will it install itself in the Registry? Or will I see the 'gear' symbol or what?

Just run the reg file and Windows will ask what to do.

[loninappleton]

Thanks, found it but made no changes yet. Once the txt is changed to reg, then will it install itself in the Registry? Or will I see the 'gear' symbol or what?

Just run the reg file and Windows will ask what to do.

Thanks and gratitude for it. It could help others well. I'm going use a backup HD to practice. My backup routine will take a bit of time to update.

[loninappleton]

Thanks, found it but made no changes yet. Once the txt is changed to reg, then will it install itself in the Registry? Or will I see the 'gear' symbol or what?

Just run the reg file and Windows will ask what to do.

So if this reg tweak works properly, all these manual installs could be avoided if that were the first step? Any guide making I do for this long Q&A would make this process a lot smoother for some users like me. I'll able to try it shortly. Or coult it be incorporated into SE's install routine or would that mean some sort of conflicts? I'm just speculating.

[VoodooFX]

Best fix would be at SE program, but probably Nikse is not so interested...
v3.6.4 is the last version when downloads worked normally on Windows 7, could be something to do with .NET version bump in v3.6.5.

[loninappleton]

@VoodooX,

That is very interesting to hear. So the Win7 download was not a problem earlier? I'll get on this with the reg tweak shortly.
One other thing: Will a backup and restore roll back any problem from the reg. tweak? I'm just very cautious about that.

[VoodooFX]

So the Win7 download was not a problem earlier?

Yes.

One other thing: Will a backup and restore roll back any problem from the reg. tweak? I'm just very cautious about that.

You don't need a backup just for a reg tweak, like you don't need a brigade of firefighters if you want to switch off the lights in your room.

Here is a "backup" reg: https://pastebin.com/pNQtms9t

Run it to revert the effect of the first reg ["Enable TLS 1.2 for Windows 7"].

[loninappleton]

I don't expect any trouble. But with all this unfamiliar turn on, turn off things, I'm making a fresh backup. Then I'll
try the reg. routine and once done, attempt to load the Whisper downloads through SE. It's just that from time to time I'd like to hear from those who've actually used these things. From SE-- I don't have it open since doing backups, is the option offered in that Video tab at audio to video to get started?

[loninappleton]

I am now on my scratch disk. Iused Revo uinstaller to take off the old install. I was trying to get the right beta but this was confusing Videohelp software shows 3.6.10/11 I put that on to the about to check the vintage. Not good, it did not look like the one shown at the link here. I am simply going to write over the 3.6.10.

Also of note, SE hangs when checking for updates in the same way as elsewhere. After I get this beta on, I'll put on the reg tweak. I'm looking for a beta called NEXT 47 in the naming convention.

[loninappleton]

I'll have two screen shots then I'll quit for the day. The short story is I'm glad I'm on a practice disk. My screen shots show the right beta from this
thread and other one shows TLS installed correctly.

With all this I had to run repair and bootrec a few times plus some other stuff to get Win7 running again. The failure to seek in check for updates in this SE Beta caused me to have to restart after hanging. I didn't go any further than that. screen shots follow....

[loninappleton]

I learned some things today.

Subtitle Edit Beta NEXT 61 which is the current link here in the thread, can access the internet and show a result for check for updates in Windows 10.

I have a Win10 disk for practice but dislike many things about using it. At some point I will see how this Win10 behaves with SE Beta loading the required things for Whisper.

Can someone say if/how the large language model for Whisper is downloaded and stored on the disk?

That is the ggml-medium-en.bin zip file. I downloaded that at one point but it might be totally wrong.

[loninappleton]

"main.exe"and "whisper.dll" needs to be in a "Whisper" folder.
And you should unpack the beta version to something like "C:\Tools\SE" - and run "C:\Tools\SE\SubtitleEdit.exe"

I overlooked this first time through and will see if I can correct for it. The naming convention not including 'Beta' has been confusing.

You are saying here Create a folder called Tools and a subfolder called SE then run The exe. for the Beta. I did miss that.

[loninappleton]

"main.exe"and "whisper.dll" needs to be in a "Whisper" folder.
And you should unpack the beta version to something like "C:\Tools\SE" - and run "C:\Tools\SE\SubtitleEdit.exe"

I overlooked this first time through and will see if I can correct for it. The naming convention not including 'Beta' has been confusing.

You are saying here Create a folder called Tools and a subfolder called SE then run The exe. for the Beta. I did miss that.

I think the Whisper will only run on my Win10 HD. But I went ahead to make this sandbox install for SE NEXT Beta 68, the newest download on the thread here. I made all the folders and subfolders at C:\Tools\Subtitle Edit Beta\subtitle edit.zip and unpacked it there. I know from the past that VOSK had to be manually installed on Windows 7. But on setting this new one up, opening VOSK at SE> Video> Audio to Video> VOSK stops at "install ffmpeg'?

I tried to put the newest ffmpeg in the folder but it is still not found. And I unzipped it to the Appdata > Roaming Subtitle edit location (there is still only one in the Roaming location, not a separate location for this beta material. (edit) Lastly, I simply copied the VOSK and Whisper data folders into the new SE Beta for any manual downloads.
I just want to get this ffmpeg done for now. So what is the proper location for ffmpeg to be seen by this new Beta install described above?

[Nikse]

The beta version are portable versions... all files should be in the folder you unpacked SE to (and sub folders), so nothing SE related should be under "Appdata".

See more here: https://www.nikse.dk/subtitleedit/help#audio_to_text

A screenshots of my folders... downloads are still not automatic? Firewall/bad-antivirus/?


[Attachment 68037 - Click to enlarge]
[Attachment 68038 - Click to enlarge]
[Attachment 68039 - Click to enlarge]

[loninappleton]

These screen shots are helpful and I'll compare that. But currently, I am not using the registry patch for TLS on/off etc. which we had above. I'll have to review the other thread where my VOSK install was successful.

edit

I see from the screen shots that all the whisper models have a folder location. One thing to do is
get those .bin files in the right location. I'll come back when I've made a bit more progress.

[loninappleton]

My update on this is I'm using my Win10 HD and the Whisper install on that including medium size meory model is going smoothly. Thanks Nikse for a well laid out install. What know-how I'll need to actually run Whisper I've yet to find out.
As a sample I have a five minute clip which is pretty much plain speaking and of course no transcript to compare.

[loninappleton]

One thing I see is that the recorded time does not refer to the time length of the clip. Processing continues well-past that.

[loninappleton]

I took the Win10 Whisper HD down after a failed attempt at Whisper. I would have to get some guidance on it. The box to open and add a WAV was not visible. I did not see any of that from what is shown using VOSK. Using the generate button on SE Whisper starts up without error and counts time of the job not the time of the content. Time of the content was 5 mins 30 secs.
I let it run 4 hours with the 'medium' model so there's a few places where I need to get some experience with it. I don't know what Whisper was decoding all that time. No errors messages were visible.

[loninappleton]

Well I mounted the Win HD with Whisper again. Today things are going better. I was not seeing something or overlooking
what I needed to see. To run the Whisper routine, I am taking the same 5:30 sec clip and extracting a WAV file for best
quality using Audacity (or Golwave.) Then I'll add that and see if the generate process works and perhaps shows the
progress bar as I usually see it.