Make anyone sound naughty with Python.
Bleep out keywords of your choice from an mp4 by leveraging a transcription model (here Whisper) to transcribe the audio, then target and replace chosen words with bleep sounds using the extracted timestamps associated with your chosen word(s).
All processing is performed locally - see the streamlit app (setup below) and detailed walkthrough notebook (see beep_that_sht_walkthrough.ipynb) to play / see nitty gritty details.
An example - below is a short clip we'll bleep out some words from using the pipeline in this repo. (make sure to turn on audio - its off by default)
bleep_test_og_cropped_low_res.mp4
Now the same clip with the words - "treetz", "ice", "cream", "chocolate", "syrup", and "cookie" - bleeped out
bleep_test_processed_cropped_low_res.mp4
To get setup to run the notebook / bleep your own videos / run the strealit demo first install the requirements for this project by pasting the below in your terminal.
pip install -r requirements.txtYou will need ffmpeg installed on your machine as well.
Start the streamlit demo
python -m streamlit run bleep_that_sht/app.pyYou will see printouts at the terminal indicating success of the two smallest whisper model downloads - tiny and base.
Then you can drag and drop your own mp4 (note: only mp4 is accepted) videos into the demo app, define your own bleep_words, and process.
You can download your bleeped video by clicking on the three dots at the bottom right of the bleeped video, and clicking download.