Hi all,

For those looking for an open-source alternative to paid captioning tools, I’ve updated a Google Colab notebook for WhisperX.

Unlike standard Whisper, WhisperX uses forced alignment, meaning the timestamps are significantly more accurate (frame-perfect in most cases). I also integrated a fix for the recent Matplotlib backend errors on Colab, so it runs smoothly.

Specs:

Model: Whisper Large-V3.

Features: VAD (Voice Activity Detection), Diarization, and Translation.

Formats: Exports to .SRT, .VTT, .JSON, .TXT.

Link to Notebook: https://colab.research.google.com/drive/1SJPTjlVlE9Je94zwtFAgOLI5zJO4yoTs?usp=sharing

Note: Make sure to enable hardware acceleration (T4 GPU) in the runtime settings for float16 precision.

Hope this helps with your post-production workflow!

Thanks. What is "Whisper X"? Another version of Whisper, or is it only the name you gave to your collab notebook?

It is actually a separate library based on Whisper (developed by m-bain, a PhD student and AI researcher at Oxford University), not a name I just made up! :)))

The main difference is that WhisperX uses 'Forced Alignment' to fix the timestamp drifting issues often found in the original Whisper. I also integrated the Speaker Diarization feature into this notebook. My work simply wraps it all in a user-friendly interface to run smoothly on Google Colab

You can check the official project here: https://github.com/m-bain/whisperX

Thank you for this alternative.

I really love the quality of your subtitles. It would be awesome to see what you can create with this tool on your next project

Thank you, I hope that's for the 14 that I actually did. :p

I have a subtitle project in progress for which I was able to find the original subtitles.
So I will use it for the next project and thus I will be able to compare Faster Whisper XXL and WhisperX to know their differences and their advantages. But I don't know when I'll do it; it could be next year or in three months.

My concerns are that my connection is not very good, and therefore the "Upload and configuration" part may cause problems if it takes too long, because Colab could disconnect me. That's why I'm using Google Drive to upload my files and link them to Colab.