video-transcriber

Extracts visually distinct frames from videos and transcribes audio using Whisper. Creates a portable zip file containing a markdown transcript with slide images in the img sub-directory.

Features

• Smart slide detection - Uses perceptual hashing to capture only distinct frames, not every frame
• Audio transcription - Uses Whisper AI locally to transcribe speech to text
• Timeline merging - Associates transcribed audio with the corresponding slides
• Portable output - Generates a zip file with markdown and images that works anywhere

Requirements

• Python 3.10+
• ffmpeg (for audio extraction)

# Ubuntu/Debian
apt install ffmpeg

# macOS
brew install ffmpeg

# Windows (using winget)
winget install ffmpeg

Installation

uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
uv pip install video-transcriber

Hugging Face

The first time you run the transcriber, it downloads Whisper models from Hugging Face. This may take several minutes.

You may see this warning:

Warning: You are sending unauthenticated requests to the HF Hub.

This is harmless. Everything will still work.

Usage

Python API

from video_transcriber.transcribe import transcribe_video

zip_path = transcribe_video("my-presentation.mp4", "output/")

The output zip file contains:

output/my-presentation_transcript.zip
├── transcript.md
└── img/
    ├── frame_000.png
    ├── frame_001.png
    └── ...

Options

zip_path = transcribe_video(
    "my-presentation.mp4",
    "output/",
    model_size="large-v3",     # Whisper model: tiny, base, small, medium, large-v3
    sample_interval=15,        # Check for new slides every N frames (default: 30)
    include_timestamps=True,   # Include timestamps in markdown output (default: False)
    audio_only=False           # Transcribe audio only, skip frame extraction (default: False)
)

Audio-Only Mode

For videos where you only need the audio transcript (podcasts, interviews, etc.):

zip_path = transcribe_video(
    "podcast.mp4",
    "output/",
    audio_only=True,
    include_timestamps=True
)

This skips frame extraction entirely, producing a zip with just the text transcript.

Development

git clone https://github.com/romilly/video-transcriber.git
cd video-transcriber
python -m venv venv
source venv/bin/activate
pip install -e .[test]
pytest

Quick Demo

Run the demo script to process the included test video:

python demo_create_zip.py

This processes tests/data/demo.mp4 and creates tests/data/generated/demo.zip.