/WebVTT-NLP

Primary LanguagePython

WebVTT NLP

This project is a tool to use NLP with Web Video Text Tracks Format (WebVTT) to batch process big Meeting Transcripts and automatically correct the transcripts using a LLM.

Overview

The main steps involved in the transcription correction process are:

  1. Parsing the Transcript: The parse_transcript function reads the input WebVTT file and extracts the UUID, timestamp, and transcript lines.
  2. Saving to JSON: The parsed data is saved to a JSON file using the save_to_json function.
  3. Preparing Messages for API Call: The prepare_messages function formats the parsed data into messages suitable for an API call to a language model.
  4. Processing Data in Batches: The process_data function processes the transcript data in batches, making API calls to correct the text.
  5. Updating JSON with Corrected Text: The parse_txt_and_update_json function reads the corrected text from the API response and updates the JSON file with the corrected lines.
  6. Converting JSON to WebVTT: The convert_json_to_txt function converts the corrected JSON data back into WebVTT format.

Usage

  1. Convert Input to JSON: Run the script to parse the input WebVTT file and save the data to input.json.
  2. Process Data: The script processes the data in batches, making API calls to correct the transcript.
  3. Update JSON with Corrected Text: The script updates the JSON file with the corrected text from the API response.
  4. Convert JSON to WebVTT: The script converts the corrected JSON data back into WebVTT format and saves it to output.txt.

Running the Script

Ensure you have a input.txt (with WebVTT contents) in the same folder as the main.py.

To run the script, execute the following command: python main.py