How To Remove Filler Words With Python [tutorial]

Published: 26 August 2023
on channel: Haseeb Kamal

1,736

Filler-words (umm, uhh, like etc.)🤬 are super annoying to listen to and also a pain to edit out in post production.
Usually, it is a manual process where I have to find the exact timestamps 🕐 for each occurrence and cut it out 🎬 in software.

Here's how you can use Python and a AI model to completely remove filler words from a video.
We use a speech-to-text model to transcribe (aka. convert speech to text) an input video. The model gives timestamps for each word uttered. Based on this information we do some Python post-processing to cut the video into chunks and merge the chunks into one final video.

The results are ... weird

We use an implementation of the OpenAI whisper model called whisper-timestamped to achieve this:
https://github.com/linto-ai/whisper-t...

Source code:
https://github.com/mhkamal1/yt-conten...

00:00 Intro
00:42 The process
01:34 Exploring speech-to-text options in Python
02:28 Whisper-timestamped
02:50 Code: Extract audio from video
03:22 Code: Load whisper model and transcribe
04:45 SUPER COOL Transcription results !!
05:41 Code: Get filler words with timestamps
07:13 Code: Split video into chunks
07:53 Code: Merge chunks into one
08:15 DEMO of Final Results!!
09:04 Closing thoughts

In short we cover:
Finding a speech-to-text model in Python
Extract audio from video
Transcribe audio
Find filler words from transcription
Split video according to timestamps
Merge chunks into one final video
How to have fun with Python

Please consider liking if you found the video useful and subscribe for more!

Subscribe here: / @haseebkamal

On this page of the site you can watch the video online How To Remove Filler Words With Python [tutorial] with a duration of hours minute second in good quality, which was uploaded by the user Haseeb Kamal 26 August 2023, share the link with friends and acquaintances, this video has already been watched 1,736 times on youtube and it was liked by 39 viewers. Enjoy your viewing!

15,503

303