Expo STT Blog

A real-time speech-to-text mobile application built with Expo and React Native. This app uses on-device AI models for real-time voice transcription without requiring internet connectivity.

Features

• 🎙️ Real-time Speech Transcription: Live voice-to-text conversion using Whisper Tiny EN model
• 📱 Cross-platform: Works on iOS and Android
• 🔒 Privacy-focused: On-device processing - no data sent to servers
• ⚡ Low latency: Optimized for real-time performance with 100ms audio chunks
• 🎛️ Audio optimizations: Configured for optimal speech recording quality

Technologies Used

• Expo (~53.0.22) - Development platform
• React Native (0.79.6) - Mobile framework
• react-native-audio-api (^0.7.1) - Audio recording and processing
• react-native-executorch (^0.5.1) - On-device AI model execution
• Whisper Tiny EN - Lightweight speech recognition model

Prerequisites

• Node.js (14 or higher)
• Expo CLI
• iOS Simulator or Android Emulator (or physical device)
• Xcode (for iOS development)
• Android Studio (for Android development)

Installation

1. Clone the repository:

git clone https://github.com/software-mansion-labs/expo-stt-blog.git
cd expo-stt-blog

2. Install dependencies:

npm install

3. Start the development server:

npm start

4. Run on your preferred platform:

# iOS
npm run ios

# Android
npm run android

Permissions

The app requires microphone permissions for speech recording:

iOS

• Microphone access is automatically requested with the message: "This app requires microphone access for real-time speech transcription."

Android

• RECORD_AUDIO - Required for audio recording
• MODIFY_AUDIO_SETTINGS - Required for optimal audio configuration

Usage

1. Launch the app
2. Grant microphone permissions when prompted
3. Tap "Start Recording" to begin real-time transcription
4. Speak into your device's microphone
5. Watch as your speech is converted to text in real-time
6. Tap "Stop Recording" to end the session

Technical Details

Audio Configuration

• Sample Rate: 16kHz (optimized for Whisper model)
• Buffer Size: 1600 samples (100ms chunks)
• iOS Audio Session: Configured for speech with playAndRecord category and spokenAudio mode

Model Information

• Uses Whisper Tiny EN model for English speech recognition
• On-device processing ensures privacy and works offline
• Supports both committed (finalized) and non-committed (provisional) transcriptions

Project Structure

expo-stt-blog/
├── App.tsx                 # Main application component
├── app.json               # Expo configuration
├── package.json           # Dependencies and scripts
├── assets/               # App icons and images
├── android/              # Android-specific files
├── ios/                  # iOS-specific files
└── node_modules/         # Dependencies

License

This project is licensed under the MIT License - see the LICENSE file for details.

Performance Tips

• Use the app in a quiet environment for best transcription accuracy
• Keep the device close to your mouth when speaking
• Speak clearly and at a moderate pace
• Ensure your device has sufficient storage and memory

Acknowledgments

• Expo - Development platform
• React Native Audio API - Audio processing
• React Native Executorch - On-device AI
• OpenAI Whisper - Speech recognition model