Dutch Pronunciation Trainer 🇳🇱

A Streamlit-based web application for A1-level Dutch learners to practice pronunciation with real-time AI feedback.

Features

• 🎯 A1-Level Content: Practice sentences organized by topics (Greetings, Food, Shopping, etc.)
• 🎧 Native Pronunciation: Listen to correct Dutch pronunciation via Google Text-to-Speech
• 🎤 Voice Recording: Record yourself directly in the browser
• 🤖 AI Feedback: Get instant feedback using OpenAI Whisper speech recognition
• 🎨 Visual Feedback: Color-coded word-by-word pronunciation analysis
• 📊 Accuracy Scores: Track your pronunciation accuracy percentage

Installation

Prerequisites

• Python 3.8 or higher
• pip (Python package installer)
• FFmpeg (required by pydub for audio processing)

Install FFmpeg

macOS:

brew install ffmpeg

Ubuntu/Debian:

sudo apt update
sudo apt install ffmpeg

Windows: Download from ffmpeg.org and add to PATH.

Install Python Dependencies

pip install -r requirements.txt

Usage

1. Start the application:

   streamlit run main.py
   

2. Open your browser to the URL shown (usually http://localhost:8501)

3. Practice:

   • Select a topic from the sidebar
   • Click "New Sentence" to get a random sentence
   • Listen to the correct pronunciation
   • Record yourself saying the sentence
   • Get instant feedback with accuracy score!

Project Structure

dutch-tutor/
├── main.py                 # Streamlit entry point
├── requirements.txt        # Python dependencies
├── README.md              # This file
├── arch.md                # Architecture documentation
├── utils/
│   ├── audio_processor.py # Audio format conversion
│   ├── tts_engine.py      # Text-to-Speech wrapper
│   └── stt_engine.py      # Speech-to-Text wrapper
├── logic/
│   ├── scorer.py          # Pronunciation scoring logic
│   └── sentences.py       # A1 practice sentences
└── assets/                # Temporary audio files

How It Works

1. Text-to-Speech: Uses gTTS to generate Dutch pronunciation audio
2. Speech Recognition: Whisper model transcribes your recording
3. Comparison: Sequence matching algorithm compares your speech to the target
4. Feedback: Color-coded results show:
   • 🟢 Green: Correct words
   • 🔴 Red: Incorrect pronunciation
   • 🟡 Yellow: Missing words
   • 🟠 Orange: Extra words

Configuration

Whisper Model Size

By default, the app uses the base Whisper model. You can change this in utils/stt_engine.py:

• tiny: Fastest, least accurate
• base: Good balance (recommended)
• small: More accurate, slower
• medium or large: Most accurate, very slow

Fuzzy Matching Threshold

Adjust pronunciation tolerance in logic/scorer.py by changing fuzzy_threshold (default: 85).

Topics Available

• Greetings
• Introduction
• Numbers (1-10)
• Family
• Food & Drink
• Shopping
• Directions
• Weather
• Colors
• Days & Time

Troubleshooting

"No module named 'audio_recorder_streamlit'"

Install the audio recorder component:

pip install audio-recorder-streamlit

Whisper model download issues

The first time you run the app, Whisper will download the model (~140MB for base). Ensure you have a stable internet connection.

Audio recording not working

• Make sure your browser has microphone permissions
• Try Chrome or Firefox for best compatibility
• Check that no other application is using the microphone

Contributing

Feel free to add more A1 sentences, improve the UI, or enhance the feedback algorithm!

License

MIT License - Feel free to use and modify for your learning needs.

Credits

• OpenAI Whisper: Speech recognition
• gTTS: Text-to-Speech
• Streamlit: Web framework
• RapidFuzz: Fuzzy string matching