How to do multiple files batch processing?

[despairTK]

Since I will modify some parameters when generating video subtitles, I need to use the command line to run it. Currently, does it support multiple videos or video files in the same folder to be processed one by one?

If batch processing is possible, can you tell me how to do it.

[Purfview]

Every command line program "supports" it.
To be precise, there is nothing to support, just write a command for your command-line interpreter to do a batch processing.

Simplest example for Command Prompt:
for %i in (*.mkv) do whisper-faster.exe --language=en --model=medium %i

This will process one by one every mkv file in a current folder.

NOTE:
It can have significant overhead as it will reload executable and model on every file, so if you have many short files you may want to process them all in one go instead of one by one.

EDIT:
From 134.5 version Whisper-Faster supports batch processing by directory, filelist or wildcard out of the box.
And there is new --batch_recursive argument.

Few usage examples:

whisper-faster.exe "D:\Clips\*.mkv" --language=en --model=medium -o=source

whisper-faster.exe "D:\Audio" --language=en --model=medium --batch_recursive

whisper-faster.exe "D:\Band\Album.m3u" --language=en --model=medium --vad_filter=False

[ghost]

does whisper-faster.exe "D:\Clips*.mkv" and whisper-faster.exe "D:\Clips" --language=en --model=medium --batch_recursive same thing?

[Purfview]

does whisper-faster.exe "D:\Clips*.mkv" and whisper-faster.exe "D:\Clips" --language=en --model=medium --batch_recursive same thing?

No. D:\Clips is path to folder.

[Purfview]

Here is Command Prompt example without overhead, process all mkv files in one go.

Save the code posted bellow as text file and rename its extension to bat, then execute it.

@echo off
cd /d "%~dp0"
set list=
for %%f in (*.mkv) do call set list=%%list%% "%%f"
call :strlen "%list%", len
exit /b

:strlen string len
SetLocal EnableDelayedExpansion
set "token=#%~1" & set "len=0"
for /L %%A in (12,-1,0) do (
    set/A "len|=1<<%%A"
    for %%B in (!len!) do if "!token:~%%B,1!"=="" set/A "len&=~1<<%%A"
)
EndLocal & set %~2=%len%

echo.
echo Batch note: Files list length is %len%.
echo.

IF %len% == 0 goto ERROR1
IF %len% GTR 8000 goto ERROR2
IF %len% LEQ 8000 goto START

:START
whisper-faster.exe --language=en --model=medium %list%
echo.
echo Batch operation complete!
echo.
pause
exit

:ERROR1
echo Error: No files found.
echo [Place .bat in the same folder where files are and check the files extension in 4th line]
echo.
pause
exit

:ERROR2
REM https://stackoverflow.com/questions/29354602
echo Error: Command is too long! [limit is 8191 characters]
echo.
pause
exit

[ClaireCJS]

You should try TCC command line, it has no limit to command line length other than your actual RAM.

I'm going to do this on 30,000 or so files.... But I want to just use the EXE since it handles things so well. I could definitely chunk it up, but it's nice to have no limitation.

[ClaireCJS]

That being said -- filelist support would be better. And it would solve the problem of length limits.

For example, an .m3u file is a playlist of mp3s/songs.

That would be great to run through this. Load the model once, and create karaoke for every song in the playlist with one go.

[Purfview]

Yeah, batch processing could be implemented into exe, maybe add some argument like --batch then supply path to folder with existing audio argument, then some code to make a list of all audio files.

Not sure if it's worth to bother when batching can be done simply by CMD, TCC & ect.
If you want to implement it then I don't mind. [cli in faster-whisper is almost same as in vanilla whisper]

[Purfview]

CMD and PowerShell have command length limit of 8191 characters, that's why there are error checks in the example in my second post.

TCC should have no limits, so below is example without error checks to run .bat in TCC:

@echo off
cd /d "%~dp0"
set list=
for %%f in (*.mkv) do call set list=%%list%% "%%f"
whisper-faster.exe --language=en --model=medium %list%
pause

[deadcoder0904]

oh, i see batch_recursive now:

whisper-faster.exe "E:\Tuts\ChatGPT Mastery" --language=en --model=medium --batch_recursive

i should probably use this from now on.

[Purfview]

Yes, these batch scripts are non needed anymore, if you don't want to do recursive then you can use -o=source.

[deadcoder0904]

thanks, i just put it in .bat file just for the sake of it.

for individual folder, use this:

subtitle-generator.bat

@echo off
cd /d "%~dp0"
set list="E:\Tuts\AI Art\ChatGPT Mastery\5 - Congratulations Next Steps"
for %%f in (*.mp4) do call set list=%%list%% "%%f"
whisper-faster.exe %list% --language=en --model=medium --output_dir=%list%
pause

for recursive, use this:

subtitle-generator-recursive.bat

@echo off
cd /d "%~dp0"
set list="E:\Tuts\AI Art\ChatGPT Mastery"
for %%f in (*.mp4) do call set list=%%list%% "%%f"
whisper-faster.exe %list% --language=en --model=medium --batch_recursive
pause

prefer terminal tbh but windows doesn't have snippet expander like mac or atleast i haven't found it.

[Purfview]

--batch_recursive doesn't work like that. Forget those batch scripts.

Here is the proper usage for your examples:

whisper-faster.exe "E:\Tuts\AI Art\ChatGPT Mastery\5 - Congratulations Next Steps\*.mp4" -l=en -m=medium -o=source

or:

whisper-faster.exe "E:\Tuts\AI Art\ChatGPT Mastery\*.mp4" -l=en -m=medium --batch_recursive

[deadcoder0904]

it worked perfectly fine since yesterday.

i used it twice.

see i did change my for loop from your .mkv to .mp4 so it worked.

although i should use your one-liner, i'd have to remember it which i don't want to so i use .bat scripts.

i just copy-paste the folder path i want subtitles on & change it in .bat file & then double-click it.

it does the job perfectly well.

[ClaireCJS]

That's exactly what I'm doing. I'll have to figure out how to do it with a filelist, but i know it's possible using only internal commands. :)

[ClaireCJS]

For what it's worth, here's a simple python pattern for taking an argument, and processing it whether it's a file, a wildcard, or a folder(recursive)

I haven't written one to take a filelist, but if the file extension isn't audio and it appears to be a filelist (easy to check, read the lines, see if any are files), it could be treated transparently in the same fashion.

def process_file(file_path, subdir):
    [main program guts here]

def process_files(file_paths):
    for file_path in file_paths:
        process_file(file_path)
    #optional ostprocess_files(".")

def process_directory(directory):
    for subdir, dirs, files in os.walk(directory):
        for filename in files:
            filepath = subdir + os.sep + filename
            process_file(filepath, subdir)
        #optional: postprocess_files(subdir)


def main():
    path = sys.argv[1]

    if os.path.isdir(path):                                                     # If path is a directory
        print(f"\nProcessing files in directory: {path}\n")
        process_directory(path)
    elif os.path.isfile(path):                                                  # If path is a file
        print(f"\nProcessing file: {path}\n")
        process_file(path, subdir=".")
    elif '*' in path:                                                           # If path is a wildcard
        files = glob.glob(path)
        print(f"\nProcessing files: {files}\n")
        process_files(files)
    else:                                                                       # If path is neither a file, nor a directory, nor a wildcard
        print("\nInvalid argument. Please enter a valid file path, wildcard, or directory.\n")

[Purfview]

It should be adapted to cli there: https://github.com/openai/whisper/blob/b91c907694f96a3fb9da03d4bbdc83fbcd3a40a4/whisper/transcribe.py#L378

[ClaireCJS]

I agree!!!!

[and that would make it so the exe you distribute supports it the, right?]

[Purfview]

Yes, it's practically same.
Check there:

whisper-standalone-win/faster-whisper cli/__main__.py

Line 220 in e0472dd

def cli():

[Purfview]

From 134.5 version it supports batch processing by directory, filelist or wildcard out of the box.
And there is new --batch_recursive argument

[ClaireCJS]

Quite awesome! I was literally showing my friends visiting last night how awesome it is and how fast it runs. I did an entire album in 4 minutes!

Other-whisper, it's like 10 minutes a song with cuda, and like 3 hours a song without.

To go from hours to minutes is A-M-A-Z-I-N-G ... thank you so much!!! That's the difference in making huge runs possible

Now i just gotta figure out splitting out the vocals with DEMUCS, pre-prompting whisper by downloading the lyrics with lyricsgenius.py/exe to reduce mistkaes, and any other options that might increase vocal accuracy and damn. It'll be a complete solution I've wanted for decades! Woo! thank you!

[Ed1ks]

Does it recognize if an .srt file with same name exists and skips it?

[Purfview]

Does it recognize if an .srt file with same name exists and skips it?

What do you mean by "it"?

[deadcoder0904]

yes, it does. there is a --skip option for the same. if you use --skip, then it will skip the same name srt file.

[gkngkngkn]

@Purfview Sir, in batch task execution(all parameters are default.), I have encountered an issue where the presence of a silent file in the batch results in a transcription error, leading to the interruption of the entire transcription process. like below

Traceback (most recent call last):
File "D:\whisper-fast_main_.py", line 977, in
File "D:\whisper-fast_main_.py", line 876, in cli
File "faster_whisper\transcribe.py", line 274, in transcribe
File "faster_whisper\audio.py", line 52, in decode_audio
File "faster_whisper\audio.py", line 103, in _resample_frames
File "faster_whisper\audio.py", line 90, in _group_frames
File "faster_whisper\audio.py", line 80, in _ignore_invalid_frames
File "av\container\input.pyx", line 202, in decode
File "av\container\input.pyx", line 134, in demux
File "av\container\streams.pyx", line 117, in av.container.streams.StreamContainer.get
IndexError: tuple index out of range
[9988] Failed to execute script 'main' due to unhandled exception!

another issue is when the transcription encounters a corrupted file that cannot be played, it also terminates the entire task.

Is there a way to generate empty subtitle files directly when meet such silent files? or automatically skip file causing transcription issues(maybe other unknow issues like a audio file is broken) and proceed to run the next file directly. to prevent the entire process from being interrupted?

hope this can be fix in the next version. thx

[Purfview]

Use --check_files switch, it will check the files for errors.

hope this can be fix in the next version. thx

You could search your error in the issues -> IndexError: tuple index out of range

[AnthonyEmertec]

It is possible in Windows to allocate global system memory which isnt attached to the calling process.

This would mean you could load the model and perform other time consuming one off initialisation on it. The first time the exe was run.

Then process and quit the whisper exe.

When it was run the next time it could use the model held in memory if it matched the next run arguments. If they arguments where different it could free the global memory and start again.

Additionally, you could add an argument to the exe to flush the global memory when finished.

This would offer the speed up of batch processing without the downsides.