Machine Learning Engineer focused on Arabic speech and NLP. I build multi-dialect Arabic text-to-speech systems and voice agents, from data and fine-tuning to serving and evaluation. Computer Engineering graduate of the Arab American University, based in Palestine 🇵🇸.
| Area | What I work on |
|---|---|
| 🗣️ Arabic TTS | Multi-dialect voices (9 Arabic dialects + Indian-accented English), built end to end: data collection/generation, cleaning, fine-tuning, serving, evaluation |
| ⚙️ Inference & serving | Fine-tuned TTS models served through a custom inference engine with hot-swappable LoRA adapters |
| 🤖 Voice agents | Deployed for banking, healthcare, education, telecom, and airports |
| 🔤 Arabic NLP | Intent extraction and routing, diacritics and pronunciation quality |
| 👥 Mentorship | Led and mentored 6 junior engineers on voice cloning, data pipelines, and fine-tuning |
NVIDIA NeMo Speech: Magpie TTS
| Contribution | |
|---|---|
| #16599 | Arabic-English code-switching for Magpie TTS: English words inside Arabic sentences go to the English tokenizer, Arabic words to the Arabic one. English words said correctly went from 0% to 91% on 150 test clips, with no retraining and no model changes |
| #16591 | Tokenizer fix: Arabic, Hindi, Italian, Vietnamese and Portuguese were silently falling back to the English tokenizer with the latest Magpie TTS checkpoint |
| #16592 | Issue: reported and measured English words in Arabic sentences being dropped or hallucinated |
| Highlight | |
|---|---|
| 📄 | Co-authored a Scopus / Web of Science indexed paper published in IJACSA |
| 🎓 | Selected as an AWS AI Future Engineer scholar (one of ~1,500 worldwide) |
| 🤝 | Amal Initiative fellow |
| 🎤 | Speaker and workshop presenter on AI/ML careers for students |
| 📚 | Udacity Nanodegrees: Intro to Machine Learning with TensorFlow and AI Programming with Python and TensorFlow |
| Topic | Focus |
|---|---|
| Arabic speech | Pronunciation and diacritics diagnosis in TTS |
| Speech models | Voice cloning, LoRA fine-tuning, evaluation of naturalness |
| LLMs | Transformers and applied LLM systems |
| Article | Platform |
|---|---|
| Mastering Image Classification: Creating a Flower Classifier with MobileNetV2 and TensorFlow | Medium |
| Where | Link |
|---|---|
| hadeel-bkhaitan | |
| Medium | @hadeel.b14 |
Always happy to talk about Arabic speech, TTS, and voice AI. 🚀