How with OpenAI Whisper do I transcribe all mp4s in a folder?

Problem:

I'd like to transcribe all mp4 files in a folder on both Mac and Windows using Whisper.

I am familiar with scripting and command-line, and have already set up Whisper to work on my machine.

Windows example solution:

The following is an example solution for Windows. Save this in a .bat file and run it in the folder you'd like to process all mp4 files. Modify the script as you need to.

This example transcribes in English with Whisper set up to use CUDA. It skips over any already transcribed mp4 files. It is not recursive.

e.g. transcribe_all_mp4.bat:

for %%f in (*.mp4) do (
	echo %%~nf
	if not exist "%%~nf.txt" (
		echo will process %%~nf.mp4
		whisper "%%~nf.mp4" --language English --device cuda
	) else (
		echo will NOT process %%~nf.mp4
	)
)

Mac example solution:

The following example is a single Terminal command you can use in MacOS with Whisper already set up.

Note that this example command does not check for existing transcriptions. It does however specify Whisper to use the medium model. It is not recursive. Modify the command as you need to.

In Terminal in the folder where you want to transcribe your .mp4 files:

for f in *.mp4 ; do whisper $f --model medium ; done

Notes:

The above solutions in Windows and Mac were verified to work between February 2023 to June 2023 (for real-life conference and film work). As of this time the good folks over at Whisper haven't yet implemented batch processing of files.

Hope these examples helped you start to get past where you too got stuck :)

References:

• https://www.digitalocean.com/community/tutorials/workflow-loop-through-files-in-a-directory
• https://github.com/openai/whisper/discussions/153
• https://stackoverflow.com/questions/39615/how-to-loop-through-files-matching-wildcard-in-batch-file
• https://learn.microsoft.com/en-us/previous-versions/windows/it-pro/windows-xp/bb490909(v=technet.10)?redirectedfrom=MSDN
• https://github.com/openai/whisper/blob/main/README.md
• https://stackoverflow.com/questions/23735282/if-not-exist-command-in-batch-file

Batch resize images to a fixed width and add watermarks

Problem:

I've got a bunch of images that I want to make a resized version of to post online. Additionally, I want to add a watermark to each of these resized images. I already have ImageMagick installed in my OS X or Linux computer.

Solution:

This is quite a specific problem with a very specific solution. This post assumes that you have read and understood the previous post found here. If not, the solution in this blog post should still be somewhat clear, if not a little fast-paced.

For the sake of this post, let's assume that the source images are a bunch of .JPG images. Additionally, let's assume that we want to resize our images to a width of 320 pixels while maintaining the aspect ratio of the original image. Finally, let's assume that we want to give our images the semi-transparent watermark "grammarofdev.blogspot.com", using the font located at "/Library/Fonts/Arial.ttf". To prevent overriding our original images, we will also create a new sub-folder called 'results'. Below is an example Terminal command of how to do all of this (with line-breaks added for legibility):

$ mkdir ./results
$ mogrify -font /Library/Fonts/Arial.ttf -pointsize 88 -verbose
          -draw "fill rgba(255,255,255,0.5) text 24,75
                 'grammarofdev.blogspot.com'"
          -path ./results -resize 320 -quality 86 -format jpg *.JPG

Phew, what an ugly, long command. What does it all mean?

The "-font /Library/Fonts/Arial.ttf -pointsize 88" selects the Arial TrueType font with a point-size of 88pt. The "-verbose" flag outputs verbose information to the terminal while the conversion is being done. The "-draw "fill rgba(255,255,255,0.5) text 30,80 'grammarofdev.blogspot.com'"" portion tells ImageMagick to render the text "grammarofdev.blogspot.com" with a fill colour of white and an alpha of 0.5, starting at the (x,y) coordinate of (24,75) of the original image.

The "-path ./results" indicates that the resized images should be written to the "results" sub-folder. Note that all files in that sub-folder with the same name will be written over without warning! Additionally, omitting this flag will write over all of your original images!

Next, "-resize 320 -quality 86 -format jpg" indicates that the images should be resized to have a width of 320 pixels, maintaining the original aspect-ratio, and then converted to jpeg with a quality of 86. Finally "*.JPG" indicates that this command should convert all images with the suffix .JPG (case-sensitive).

Note that the text is drawn onto the original image prior to the image being scaled down to the final desired size.

Let's see how well this worked. The following two images will show an image before this command is run, and the resulting image afterwards. Note that the original image dimensions are 3264x2448 pixels, however this image is not included in this post. (A slightly-larger version can be found here!)

Original image (scaled to width 300px):

Watermarked image:

Where can this be useful:

Imagine that you had a hundred images you wanted to automatically watermark and resize, or imagine that you have created a web app that requires re-sizing and watermarking previews of images. Using ImageMagick to perform this task is a quick way to achieve the desired results without needing to program anything complex, or use programs such as Photoshop. This post only touched the surface of what ImageMagick can do. For more details on resizing, or details on drawing on images automatically with ImageMagick, check out the following references.

References:

• http://www.imagemagick.org/www/command-line-options.html#draw
• http://dptnt.com/2007/09/batch-resize-images-using-imagemagick/

Batch convert PNG images to 50% scaled jpg images with mogrify

Problem:

How do I convert a bunch of PNG images to jpeg images at 50% of their original dimensions? I'm on an OS X computer and have already installed ImageMagick.

Solution:

You can use the 'mogrify' command that comes with ImageMagick in the Terminal. Note however, that mogrify has the ability to silently overwrite images, so *be careful*! In the following example a destination folder is created to save images in order to preserve the original images.

$ mkdir ./results
$ mogrify -path ./results -resize 50x50% -quality 80 -format jpg *.PNG

So what did all of that mean?

The first part, "mkdir ./results" created a sub-folder named 'results' to save all of the converted images. You can make a different sub-folder if you already have a folder named 'results' to avoid overwriting contents in that existing sub-folder. Remember to change the folder in the -path portion of the mogrify command, too.

The next part batch converts all of the .PNG images to .jpg images of half the width and height. Note that the extensions are case-sensitive. The "-path ./results" flag indicates that results of the conversion are written to the "./results" folder. All destination images are overwritten if they already exist. The "-resize 50x50%" portion shrinks the final width and height to 50% of the original. The "-quality 80" indicates the compression level of the resulting jpeg images. The "-format jpg" indicates that the final image format should be jpeg. Finally, the "*.PNG" indicates that the input images will be every .PNG image in the current folder.

References:

• http://linux.die.net/man/1/mogrify