--- title: Arabic Speech to Route emoji: 🗺️ colorFrom: indigo colorTo: blue sdk: docker python_version: "3.10" app_file: speech_recognition_api.py pinned: false --- # 🗺️ Arabic Speech-to-Route (Live Demo) This Space demonstrates an Egyptian Arabic NLP routing extraction pipeline. It takes voice or text input in highly colloquial Egyptian Arabic (e.g., "عايز اروح التجمع الخامس من الهرم") and extracts the origin and destination using a hybrid NER (Named Entity Recognition) + Gazetteer pipeline. ## How it works 1. **ASR:** Whisper (small) fine-tuned on Egyptian Arabic transcribes the audio. 2. **Normalization:** Custom text cleaning handles orthographic noise (e.g., Alif/Hamza variants). 3. **NER:** A Transformer model extracts spatial intent. 4. **Gazetteer:** A fuzzy phonetic matching system resolves the entities against a known location database. For source code, API references, and architecture details, please visit the GitHub repository.