📚 Localized Audio Learning from Any Textbook

Github : Smart-Audio-Textbook

🚀 Inspiration

Many students across India and beyond face challenges in accessing educational content — whether due to language barriers, visual impairments, or simply a preference for auditory learning. While audiobooks exist, they're limited in language diversity and availability. We wanted to create a solution that could turn any textbook, in any format (PDF or image), into an audio-based learning experience — localized in the listener's language.

💡 What it does

Smart Audio Textbooks is a Streamlit web app that:

• Accepts textbook input as scanned images or PDFs.
• Extracts and sanitizes the text using OCR.
• Translates the content into regional Indian languages (like Hindi, Marathi, Tamil, etc.).
• Converts the translated text into audio using Google Text-to-Speech (gTTS).
• Plays the audio directly within the app for an instant learning experience.

🛠 How we built it

Tech Stack:

We modularized the code with a clean utils folder, handling:

• OCR (ocr.py)
• Sanitization (sanitizer.py)
• Chunking (chunker.py)
• Translation (translator.py)
• Audio generation (texttovoice.py)

🧗‍♀️ Challenges we ran into

• gTTS generated robotic voices, reducing the natural feel of the narration.
• Handling large PDFs led to performance issues due to translation and audio processing limits.
• Translating content into the same language (e.g., English to English) wasted time — no input-output language checks were in place.
• The initial chunking and text cleanup logic were too naive for complex documents.
• OCR sometimes introduced noise that made translation inaccurate.

🏆 Accomplishments that we're proud of

• Successfully built an end-to-end working MVP within a short span.
• Modular utility-based design for easy iteration and future improvements.
• Seamless support for multiple Indian languages — all offline-friendly post-setup.
• First-time Streamlit and OCR integration experience went smoothly!

📚 What we learned

• Working with scanned PDFs is more complex than it seems — OCR is sensitive to formatting and quality.
• Translating in chunks ensures better audio output and reduces failures.
• Streamlit makes building data/AI tools super accessible, even for beginners.
• gTTS is useful but limited; exploring better TTS solutions (like Coqui or Azure TTS) could be a game-changer.

🔮 What's next for Localized Audio Learning from Any Textbook

• ✅ Add language comparison to skip redundant translations.
• 🧠 Use better chunking logic for natural sentence splits and flow.
• 🗣️ Integrate better, more human-sounding TTS options (like Coqui or ElevenLabs).
• 📚 Enable flashcard or summary generation for revision.
• 🌐 Explore mobile-first design for rural learners.
• 🛡️ Add error handling, session storage, and performance enhancements.

Built With

• googletrans
• gtts
• nltk
• python
• streamlit
• tesseractocr