Transcribe. Summarize. Locally.

LeteScribe is a simple Python-based application that transcribes and summarizes automatically your videos, locally.

It uses Whisper for automatic video transcription and Ollama for summarizing it.

To run LeteScribe, you will need:

• Python 3.8+
• Whisper package and model: pip install -U openai-whisper
• Ollama: download on your computer and download a model from the model library.

To use LeteScribe, simply run the letescribe.py script with the following command:

python src/letescribe.py --language es --media-file <video_file_path>  --summarize-locally

Replace <video_file_path> with the path to your video file. The options available are:

• --transcribe-only: Transcribes the video only, skipping summary generation.
• --language: Specify the language of the video (default: English).
• --media-file: Path to the video file. It is required, if forgotten, it will ask you for the video path.
• --summarize-locally: Summarizes using Ollama API (locally).

Measured in a HP EliteBook 660 G11 (with Intel(R) Core(TM) Ultra 7 165H @1.40 GHz, 32 GB of RAM and running Windows 11):

• Video length: 2056 seconds (17:36).
• Transcription took 553.94 seconds (9:13).
• Summary took 461.80 seconds (7:41)

For any issues or errors encountered while using LeTescribe, please refer to the following:

• Check if your video file is in .mp4 format. I believe Whisper accepts more formats, including audio formats, but .mp4 was enough for my usage.
• Verify that you have the correct Whisper model loaded (turbo by default).
• Ensure that Ollama is running and you have the correct model loaded (llama3.1:8b by default)
• Write an issue on this repo.