IEEE Publication · Major Project · Aug 2023 – Jan 2024

PIB
Multilingual Press-Release Dissemination Pipeline

Python DistilBART Selenium + BeautifulSoup PyMuPDF GoogleTranslate gTTS Blender NLP / ML
View on GitHub Read the paper (IEEE Xplore)

About the project

The Press Information Bureau (PIB) publishes high-volume daily press releases that are crucial for public awareness, but long-form government text goes mostly unread. This project makes PIB content accessible by automatically collecting each release, summarising it with an abstractive transformer, translating it into Indian regional languages, and converting it to text-to-speech audio — so updates can be consumed quickly, in one's own language, and by visually impaired users.

Published as a peer-reviewed paper at the 5th International Conference on Innovative Trends in Information Technology (ICITIIT 2024), IIIT Kottayam (IEEE, 15–16 March 2024): "Automated Multilingual Multimedia Dissemination of Government Press Releases" — DOI 10.1109/ICITIIT61487.2024.10580027.

What I built / how it worked

  • Collect — automated web crawling with Selenium + BeautifulSoup: scrapes pib.gov.in ministry release listings, maintains per-ministry Excel trackers, and auto-downloads PDFs into organised folders.
  • Extract — PDF → structured JSON with PyMuPDF: id, ministry, title, date, location, main_content via text extraction and regex parsing.
  • Summarize — abstractive summarization with DistilBART (distilbart-cnn-12-6), with length-aware rules (≥800 words skipped, ≤130 words passed through).
  • Translate — English + 9 Indian languages (Hindi, Urdu, Gujarati, Marathi, Telugu, Kannada, Malayalam, Tamil, Bengali), each prefixed with a localized greeting; design targets 12 languages.
  • Speak — gTTS text-to-speech per language, opening government news to auditory consumption and visually impaired users.
  • Multimedia — designed a Blender 3D-mascot stage with Python-driven lip-sync for audio-visual clips.

Feature highlights

  • End-to-end runnable pipeline (pipeline/) with 41 sample press releases shipped in the repo.
  • Per-release multilingual output structure: translated JSON + Audio_Files/*.mp3 per language.
  • IEEE-co-sponsored conference publication — system design accepted and published.
← Back to all projects