README.md

metadata

title: Arabic Pronunciation Checker
emoji: 🎤
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.49.0
app_file: app.py
pinned: false
license: mit

Arabic Pronunciation Checker

An interactive tool for checking Arabic pronunciation using a fine-tuned Wav2Vec2 model.

Features

• 🎤 Real-time Audio Recording: Record yourself directly in the browser
• 📝 Sample Sentences: Pre-loaded Arabic sentences with reference phonemes
• ✓ Instant Feedback: Get immediate pronunciation analysis
• 📊 Detailed Results: See which phonemes you pronounced correctly and which need improvement
• 🎯 Accuracy Score: Overall pronunciation accuracy percentage

How to Use

1. Choose a sentence: Select from sample sentences or enter your own Arabic text
2. Add reference phonemes: Provide the correct IPA phonetic transcription
3. Record audio: Click the microphone button and read the sentence
4. Get results: Click "Check Pronunciation" to see your analysis

Model

This space uses the Wav2Vec2-mispronunciation-detector model, which is based on:

• Pre-trained Wav2Vec2 encoder (elgeish/wav2vec2-large-xlsr-53-arabic)
• Cross-attention mechanism for audio-phoneme alignment
• CTC loss for sequence prediction

Example

Input Sentence: السلام عليكم
Reference Phonemes: a s s a l aa m u ʕ a l aj k u m
Your Recording: [Audio]
Results: Accuracy score + detailed phoneme-by-phoneme comparison

Technical Details

• Architecture: Wav2Vec2 + Transformer blocks with cross-attention
• Training Data: IqraEval dataset
• Input: 16kHz audio
• Output: IPA phoneme sequence

Citation

@misc{wav2vec2_arabic_pronunciation,
  title={Wav2Vec2 Arabic Mispronunciation Detector},
  author={Haitam},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/Haitam03/Wav2Vec2-mispronunciation-detector}
}