Easy to use Jupyter Notebook for Youtube video inference 📺

[ArthurFDLR]

I've made a simple Jupyter Notebook including Colab forms to ease Whisper inference on Youtube videos and save the result on your Google Drive.
This is mainly meant for non-technical folks, but the parameter selection GUI is also very useful for more advanced use cases and fine-tuned inference experimentation.

[dgoryeo]

Hi Arthur,

Great job on mking additional parameters available (in the UI). Can this be used for local files too?

Thanks

[dgoryeo]

Quick update: mounting Google Drive works well. Also, I was able to run the script on an mp3 file. Unfortunately, although I had selected srt output, it only produced plain text --no timestamps.

The next task for me is to learn all the parameters. Thanks!

[ArthurFDLR]

Thanks for the input. I fixed the issue with the .srt outputs; there was a typo in my code.

[dgoryeo]

Thanks. I'll give it a try.

[dgoryeo]

Hi, I think I am missing a settings in my run. I just tried the notebook on a long (local file) video, ca 2hrs long. I did not change any of the parameters and I used medium model. The result was quite poor. The transcription output was repetition of a single line throughout the transciption. The single line was detected at 4th second of the video and then repeated for 493 times over 2 hours. There must be a paremeter/flag/setting that I have gotten wrong but I can't figure out which one.

[Abdurrafey-Siddiqui]

@ArthurFDLR
How can I transcribe multiple vids? Can you add that to the colab code please.

[ghost]

can you add playlist transcript please?

[ArthurFDLR]

Hey @Venomous912 and @toprak, thanks for the input. I just updated the Notebook to support batch processing:

• All videos within a Youtube playlist;
• All video/audio files in a Google Drive folder;

[ghost]

thank you so match

[dgoryeo]

Have you considered adding a VAD to this --like Silero?

[ghost]

Also, naming output as same as videos would be great

[MaximeRivest]

you might like jupyter-whisper :D