Transcribe Large Audiofile with Whisper

Setup

1. Require Python 3.10 and FFMPEG installed
2. Recommended (why to use venv) Create venv and enter in venv

   python -m venv venv
   
   # Linux
   . venv/bin/activate
   # Windows
   . venv/Script/activate

3. Install requirements

   # Make sure the (venv) is flag in the terminal
   pip install -r requirements.txt

4. Fix ffmpeg install if failures : Uninstall and reinstall ffmpeg-python

   pip uninstall ffmpeg
   pip uninstall ffmpeg-python
   
   pip install ffmpeg-python
   

Run

With the venv activated.

Usage:

python app.py stuff.mp3

Result will be stored in stuff_transcribe_result.txt

• Check the demo

Run with model param

• Model size for whisper are those : "tiny", "small", "base", "medium", "large"
• The argument for the model is in 2nd
• By default the large model will be used

Run on tiny model:

python app.py stuff.mp3 tiny

Run on base model:

python app.py stuff.py base

Logic

• Moviepy will chunk the mp3 audio file into smaller chunk audio
• The audio chunks are then stored in ./tmp_chunks_audio_speach2text/
• For each chunk audio (in the right order), execute the whisper model (60 seconds maximum)
• Append all text result
• Write to file result stuff_transcribe_result.txt
• Print result to stdout

Once the process is done, you can clean tmp_chunks_audio_speach2text/ folder