Whisper AI is a powerful speech-to-text model by OpenAI that allows for high-quality transcription. This guide walks you through the step-by-step installation process.

Step 1: Install Python

Whisper AI requires Python to run.

• Download Python from python.org.
• Ensure you install Python 3.8 or later (Whisper supports up to Python 3.11).
• During installation, check the box to add Python to PATH.
• Verify installation by running: python --version

Step 2: Install PyTorch

Whisper AI depends on PyTorch for deep learning functionalities.

• Visit pytorch.org and follow the instructions for your system.
• Example installation command: pip install torch torchvision torchaudio
• Verify installation: python -c "import torch; print(torch.__version__)"

Step 3: Install Chocolatey (Windows Users Only)

Chocolatey is a package manager for Windows.

• Open PowerShell as an administrator and run: Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString('https://community.chocolatey.org/install.ps1'))
• Verify installation: choco --version

Step 4: Install FFmpeg

FFmpeg is required for handling audio files.

• Windows (via Chocolatey): choco install ffmpeg
• macOS: brew install ffmpeg
• Linux (Ubuntu/Debian): sudo apt update && sudo apt install ffmpeg
• Verify installation: ffmpeg -version

Step 5: Install Whisper AI

• Run the installation command: pip install -U openai-whisper
• Alternatively, install directly from GitHub for the latest version: pip install git+https://github.com/openai/whisper.git
• To update Whisper AI: pip install --upgrade --no-deps --force-reinstall git+https://github.com/openai/whisper.git
• Verify installation: whisper --help

Step 6: Run a Test Transcription

To check if Whisper AI is working, run:

whisper example.mp3 --model small

This will generate a transcription of example.mp3.

Additional Features

• Use Different Models: Whisper supports multiple models (tiny, small, medium, large). Example: whisper example.mp3 --model medium
• Transcribe Multiple Files: whisper file1.mp3 file2.mp3
• Specify Language: whisper example.mp3 --language English
• Translate Non-English Speech to English: whisper example.mp3 --task translate

CUDA Compatibility for GPU Acceleration

If you have an NVIDIA GPU, you can speed up transcription with CUDA:

• Install CUDA-compatible PyTorch by selecting the correct version from pytorch.org.
• Install NVIDIA drivers and CUDA from NVIDIA.
• Run Whisper using CUDA: whisper example.mp3 --model large --device cuda

Congratulations! 🎉 You have successfully installed and set up Whisper AI for transcription. For further details, visit the Whisper GitHub repository.

Command:

ffmpeg -i "PresentationVideo.mp4" -filter_complex "select=gt(scene\,0.2)" "slides/%04d.jpg" -vsync vfr

Explanation of Each Component:

1. ffmpeg

• The command-line tool used for video and audio processing.

2. -i "PresentationVideo.mp4"

• Specifies the input video file: PresentationVideo.mp4 (the recorded PowerPoint or slide presentation).

3. -filter_complex "select=gt(scene\,0.2)"

• -filter_complex: Enables complex filtering.
• select=gt(scene,0.2):
  • Uses the scene detection filter to extract frames when significant slide transitions occur.
  • scene is a built-in metric that detects the difference between consecutive frames.
  • gt(scene,0.2):
    • gt() (greater than) selects frames where the scene change metric exceeds 20% (0.2).
    • This ensures that only major slide transitions are captured, avoiding minor visual changes.

4. "slides/%04d.jpg"

• Saves the extracted slides as .jpg images in the slides/ directory. You must create this directory before using the command, or it will not work
• %04d ensures images are numbered sequentially (e.g., 0001.jpg, 0002.jpg).

5. -vsync vfr

• Ensures that only variable frame rate (VFR) frames matching the scene filter are kept, preventing unnecessary duplicate frames.

Effect of Changing the Scene Threshold (scene value)

Example with a Lower Scene Change Value

ffmpeg -i "PresentationVideo.mp4" -filter_complex "select=gt(scene\,0.1)" "slides/%04d.jpg" -vsync vfr

• This command captures more frequent slide changes, useful if the slides have animations or frequent minor transitions.

Key Takeaways:

• Lower scene values (e.g., 0.05–0.1) → Extract more frames, including small slide changes.
• Higher scene values (e.g., 0.2–0.3) → Extract fewer frames, focusing on major slide transitions.
• Adjust the value based on the type of PowerPoint presentation (animated vs. static slides).