Whisper and its many forms
- Thread starter SamKook
- Start date
-
Akiba-Online is sponsored by FileJoker.
FileJoker is a required filehost for all new posts and content replies in the Direct Downloads subforums.
Failure to include FileJoker links for Direct Download posts will result in deletion of your posts or worse.
For more information see this thread.
Just press "Show code" for the "Setup Whisper" part and enter it in there:
... looking at that screenshot, they were already theoretically installed, I guess the whisper install messes them up so the original line with them could just be moved to the end instead.
Edit: It seems that spleeter is causing the problem(and preventing the rest of the packages on the line to be installed). Simply deleting spleeter seems to work fine, which makes the code look like this:
Not sure why or if it's important, but I get an srt with just deleting that.
... looking at that screenshot, they were already theoretically installed, I guess the whisper install messes them up so the original line with them could just be moved to the end instead.
Edit: It seems that spleeter is causing the problem(and preventing the rest of the packages on the line to be installed). Simply deleting spleeter seems to work fine, which makes the code look like this:
Not sure why or if it's important, but I get an srt with just deleting that.
Last edited:
Okay, it worked out! Thanks! This is old, the result is not like the latest changes, synchronisation problems, and so on, but it's still something until it's fixed.
Last edited:
Hi Sam, I tried deleting the spleeter and that didn't work, neither did adding the line you provided.
I could be doing something wrong, but I've tried, deleting the word "spleeter", and then deleting the line that "spleeter" is on, and then adding the line you quoted, none of that worked.
mei2 fixed the VADpro one so just use that instead.
If you don't want to, I'd need to know what error you're getting.
If you don't want to, I'd need to know what error you're getting.
Hello mei2 and SamKook, I've used mei2's updated VADpro, I'm getting the following error:
"The error message NameError: name 'source_separation' is not defined indicates that the variable source_separation is being used in your code before it has been assigned a value. Looking at your code, it seems likely that the source_separation variable was intended to be defined in the "Whisper Transcription Parameters" section, but it is missing. This causes the error to occur when the code reaches the line if source_separation: as it doesn't know what source_separation refers to."
I'm uploading the individual audio files and copying them into the audio path.
I really appreciate all the help!
"The error message NameError: name 'source_separation' is not defined indicates that the variable source_separation is being used in your code before it has been assigned a value. Looking at your code, it seems likely that the source_separation variable was intended to be defined in the "Whisper Transcription Parameters" section, but it is missing. This causes the error to occur when the code reaches the line if source_separation: as it doesn't know what source_separation refers to."
I'm uploading the individual audio files and copying them into the audio path.
I really appreciate all the help!
That means it doesn't see the variables declared in the first block "Whisper Transcription Parameters" so just make sure you execute it. Try executing it again if you did but still got the error.
If that fails, you can read from here how to get around that particular problem by copying that code block elsewhere: https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4860978
If that fails, you can read from here how to get around that particular problem by copying that code block elsewhere: https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4860978
The only difference is the first block that you need to execute to save the options you choose(you can also just leave it all on default) so press the play icon on that as you do for all the other block of codes.
The rest works the same as the original. If you still have trouble that my answer to DocNic above doesn't fix, then just provide the error message here and we'll see how to fix it.
The rest works the same as the original. If you still have trouble that my answer to DocNic above doesn't fix, then just provide the error message here and we'll see how to fix it.
My problem is uploading the audio file via google drive, somehow it dont work for me, i by making a folder for the same field. Maybe I'm wrong somewhere, but it's not working. Where exactly is this code block put?
Last edited:
Where does it differ when following the tutorial on how to upload using google drive in the first post of this thread? I've only ever used that option to make the tutorial so maybe it changed since or you're running into a problem I didn't.
You don't need to put the code block anywhere, it's just there at the top of the pro colab and you have to make sure to execute it(which I've been guilty of forgetting to do).
Unless you mean on the original colab, then just look at the screenshot on this post: https://www.akiba-online.com/threads/whisper-and-its-many-forms.2142559/post-4870044
That was just some quick debugging late last night and I was posting as I was testing.
What mei2 did on the pro version is the correct thing, to put spleeter on its own line, that way it doesn't get in the way if it fails and if it works, great.
You don't need to put the code block anywhere, it's just there at the top of the pro colab and you have to make sure to execute it(which I've been guilty of forgetting to do).
Unless you mean on the original colab, then just look at the screenshot on this post: https://www.akiba-online.com/threads/whisper-and-its-many-forms.2142559/post-4870044
That was just some quick debugging late last night and I was posting as I was testing.
What mei2 did on the pro version is the correct thing, to put spleeter on its own line, that way it doesn't get in the way if it fails and if it works, great.
Last edited:
It has different default settings than the other colab.
If you look at the notes on the bottom:
So better default for quality but the tradeoff is it's slower.
The test I did took 1h53m for a 3 hour movie leaving it all on default.
If you look at the notes on the bottom:
Code:
Settings for large-v2 higer quality output
(Note 1: it increases the execution time)So better default for quality but the tradeoff is it's slower.
The test I did took 1h53m for a 3 hour movie leaving it all on default.
I tried deleting "spleeter" from the code before I ran it, but still no joy. Do you have the original setup code?
To be clear, the deleting spleeter fix is for the original WhisperWithVAD by ANonEntity and not the maintenance release by mei2 which people have been using for a long time by now. That maintenance release has an additional issue of forcing a too old torch version so it needs additional modifications.
Since mei2 has fixed the pro version, you can simply use that or copy the "Setup Whisper" code from there and then replace any of the other version "Setup Whisper" code with it to fix them.
Don't forget to press the play icon after you do a modification to run the code again or it will have no effect.
If you don't want to copy the code, you can also fix them directly these ways:
You will see the errors when executing the "Run Whisper" code block but all modifications need to happen in the "Setup Whisper" code block.
To fix a missing "ffmpeg", "srt" or "deepl" module error when executing the last block, put spleeter on its own line(applies to both the original and maintenance release):
Meaning transform this line of the "Setup Whisper" code block
into
To fix a missing "torch" module error when executing the last block, remove the specific version for torch, triton and tiktoken(applies to the maintenance release).
Meaning transform this line
into this line
Since mei2 has fixed the pro version, you can simply use that or copy the "Setup Whisper" code from there and then replace any of the other version "Setup Whisper" code with it to fix them.
Don't forget to press the play icon after you do a modification to run the code again or it will have no effect.
If you don't want to copy the code, you can also fix them directly these ways:
You will see the errors when executing the "Run Whisper" code block but all modifications need to happen in the "Setup Whisper" code block.
To fix a missing "ffmpeg", "srt" or "deepl" module error when executing the last block, put spleeter on its own line(applies to both the original and maintenance release):
Meaning transform this line of the "Setup Whisper" code block
Code:
!pip install deepl srt ffmpeg-python spleeter
Code:
!pip install deepl srt ffmpeg-python
!pip install spleeterTo fix a missing "torch" module error when executing the last block, remove the specific version for torch, triton and tiktoken(applies to the maintenance release).
Meaning transform this line
Code:
!pip install --no-cache-dir torch==1.12.1 torchvision torchaudio torchtext torchdata triton==2.0.0 tiktoken==0.3.3
Code:
!pip install --no-cache-dir torch torchvision torchaudio torchtext torchdata triton tiktoken
Last edited:
Regular whisper takes about 15 minutes per hour of video on my computer. How much faster would other versions be and how much worse are the subs they make? I'd like to speed things up a bit but not if it means a big loss of quality.
Edit: This is with the large V2 model, not sure whether some models work better with some versions of whisper or not.
Edit: This is with the large V2 model, not sure whether some models work better with some versions of whisper or not.
Last edited: