transcript-pl

transcript-pl is a Node.js application designed to process document files (such as PDFs) by converting them into images and extracting text using OCR technologies (Tesseract and Google Vision). It supports AI-based transcription and can be configured for multiple languages.

Features

• PDF to image conversion
• OCR with Tesseract and Google Vision
• AI-powered transcription (OpenAI GPT)
• Configurable via YAML file
• CLI interface for flexible usage

Setup

1. Clone the repository

git clone <repository-url>
cd transcript-pl

2. Install dependencies

npm install

3. Download Tesseract traineddata files

npm run setup:tessdata

This script downloads the required .traineddata files for Tesseract OCR into resources/tessdata.

4. Configure the application

Edit the configuration file at src/config/config.yaml to set your API keys, languages, and other options.

• For Google Vision, set the path to your API key JSON file.
• For OpenAI, set your API key and model.

Usage

Run the application from the command line:

node src/app.js --document <path-to-document> [options]

CLI Arguments

Argument	Alias	Type	Description	Required
--document	-d	string	Path to the document file to process	Yes
--pages	-p	string	List of pages to process via AI and image generation (e.g., 1,3,5)	No
--lang	-l	string	Source language code for the document (e.g., pl, en, fr). Used for all OCR engines.	No
--target-langs	-t	string	Comma-separated list of target languages for AI translation (e.g., fr,en,de)	No
--example	-e	string	Path to an example file to improve AI transcription	No
--docType	-D	string	Type of document (e.g., mémoire historique, acte de naissance). Used for AI prompt.	Yes

Example:

node src/app.js --document input/sample.pdf --pages 1,2,3 --lang pl --target-langs fr,en --docType "mémoire historique"

Configuration

All main settings are in src/config/config.yaml:

• lang:
  • source: language code of the document (used for OCR, required)
  • target: array of language codes for AI translation (optional, can be empty)
• docType: Not in YAML! Always provide via CLI.
• ai: Enable/disable AI, set OpenAI API key, model, etc.
• google_vision: Enable/disable, set API key path.
• tesseract: Enable/disable.
• pdf_to_image: Set DPI for image conversion.

How it works

• The PDF is converted to images, and each page is processed by all enabled OCR engines (Tesseract, Google Vision).
• For each page, the AI (OpenAI GPT) improves the text in the source language using all available OCR transcriptions.
• If target languages are specified, the improved text is then translated by the AI into each target language.
• Result:
  • If no target language is set, you get only the improved text in the source language.
  • If target languages are set, you get both the improved source text and AI translations for each target language.

Scripts

• npm run setup:tessdata — Downloads Tesseract .traineddata files for supported languages.

Dependencies

• @google-cloud/vision
• tesseract.js
• openai
• js-yaml
• yargs
• sharp
• pdf2image
• image-size
• follow-redirects

---

Note: Make sure you have the required API keys for Google Vision and OpenAI before running the application.