LangLearnDemo is a phonetic emergency sentence learning app designed to help users overcome language barriers in India by teaching them how to pronounce essential phrases correctly in regional languages.
India has 22 official languages and hundreds of regional dialects.
Travelers, migrants, and citizens in unfamiliar states may face life-threatening situations if they cannot communicate basic needs or distress.
This app addresses that by helping users speak and pronounce critical emergency sentences clearly — even if they don't understand the script.
- 🔤 Select your known language: English
- 🗣️ Learn how to speak Hindi, Marathi, or Kannada emergency phrases
- 📖 View full-sentence phonetic translations for clear pronunciation
- 📶 Works offline — lightweight and accessible
- 🎯 Designed to integrate AI feedback and emergency detection in future versions
- 🔐 Secured with phone number + OTP login (learning app) and email-authenticated upload (recording app)
⚠️ Note: This demo includes only 2 target languages and a limited set of 7 emergency sentences to prevent replication.
The full app(learning app) currently supports 10 Indian languages and includes 51 curated emergency sentences, with more being added gradually.
The goal is to expand coverage to all 22 official Indian languages in future releases.
| Section | Screenshot |
|---|---|
| Language Selection Which language do you want to learn? |
 |
| Sentence List English Interface |
 |
| Phonetic Learning Word-by-word pronunciation (Marathi from English) |
 |
👉 To view all demo screenshots, visit the demo-app/ folder.
This private companion app is used to collect native speaker recordings securely and efficiently.
While the APK and source code are not shared due to privacy concerns, the following screenshots illustrate the complete data collection workflow.
| Step | Screenshot |
|---|---|
| Login with Email |  |
| Sentence to be Spoken |  |
| Choose Language |  |
| Mic Popup for Recording |  |
| Choose Gender |  |
| Send to Firebase |  |
⚠️ The recording app is used strictly for private data collection. Voice recordings are anonymized and securely stored in Firebase. This section is included for transparency with reviewers and collaborators.
## 🔍 Project Structure ``` LangLearnDemo/ ├── app/ # Source code of the Android demo app │ └── demo/ # Android Studio project files │ ├── demo-apk/ │ └── app-debug.apk # Pre-built APK for direct testing │ ├── assets/ │ ├── demo-sentences.json # Sample JSON used in phonetic learning │ └── screenshots/ │ ├── demo-app/ # Demo app screenshots (UI walkthrough) │ └── recording-app/ # Recording app screenshots (private app) │ ├── ai-model/ # AI architecture, pipeline, and progress │ └── README.md │ ├── LICENSE # Open-source MIT license └── README.md # Main project documentation ```
I am collecting data and training a custom AI model using real voice recordings from native speakers of various Indian languages. This model powers two key features of the app:
- Uses MFCCs, TensorFlow, and Librosa to analyze speech input
- Compares user pronunciation to validated native recordings
- Provides real-time feedback — users must reach ≥75% accuracy to progress
- Focused on low-resource Indian languages and phonetic correctness
- Analyzes real-time or uploaded audio
- Converts speech to text using custom STT models
- Flags distressing or dangerous phrases via NLP-based intent classification
- Alerts the user with on-screen phonetic guidance and optional Bluetooth voice prompt
📂 Full pipeline documentation → ai-model/README.md
-
🔐 The recording app uses email/password Firebase Authentication with device-based access control.
Only pre-approved emails (stored in Firestore) can upload recordings. -
📱 The learning app is secured using Firebase Phone Authentication with OTP verification, ensuring real-user access for mobile learning.
-
🔐 Both apps have strict Firebase rules to prevent unauthorized read/write access.
Authenticated users can only write to specific paths — all data is anonymized and write-only. -
🚫 No personal identifiers (e.g., names, IPs, locations) are attached to recordings.
-
📂 Firebase credentials and storage buckets are kept secure and are never exposed in the public repo or APK.
✅ Security features were implemented to protect user data and ensure privacy in both apps, using Firebase’s authentication and access control tools.
- ✅ Phonetic breakdown UI for sentence-based language learning
- ✅ Language-linked JSON structure for multilingual support
- ✅ Firebase-secured voice data collection pipeline
- 🟡 AI pronunciation scoring model — training in progress
- 🟡 Emergency phrase detection model — dataset tagging and early training
- ⏳ Offline-first version with TTS playback and real-time pronunciation scoring
If you're a native speaker of an Indian language and want to contribute authentic pronunciation samples or help build multilingual AI tools, feel free to reach out:
📧 Email: yadav.akanksha03@gmail.com
This project is licensed under the MIT License.
You may use or modify this project with attribution. Redistribution of collected datasets or trained models is strictly prohibited.