From published research to production-deployed AI — TTS · Prosody Control · Voice Conversion · Applied NLP
🎯 Open to: ML/AI Engineer & Applied Research roles in speech, audio and NLP — France / Remote-friendly
3 published papers (ICNLSP, JEP, EUSIPCO), open-source models on Hugging Face, and a production RAG assistant live in front of HEC Paris users — I don't just publish research, I ship it.
Live, in-use application — not a prototype. Retrieval-augmented assistant combining document retrieval with an agentic LLM pipeline, deployed as a full web app on Azure, serving real HEC Paris users.
Impact: robust tool-calling orchestration + cost-efficient inference at production scale.
Live app (HEC login required) · Stack: RAG · LLM orchestration · vector retrieval · Flask · Azure deployment
Controllable French speech synthesis with explicit SSML planning for pauses, timing, and emphasis — symbolic pause planning, break prediction, SSML generation, reproducible training/inference.
Repository · Demo · Paper — ICNLSP 2025 · HF: text2breaks · break2ssml
Waveform resynthesis from frozen WavLM representations: adversarial reconstruction (HiFi-GAN + MPD/MSD), learned weighted layer fusion over WavLM-Base+, chunked overlap-add inference, layer-ablation studies. Trained on 238h of cleaned French speech (SIWIS, M-AILABS, Common Voice).
Repository · Demo · HF Models · Paper — JEP 2026
Component-level adaptation of CosyVoice2 for European languages (French & German): language-specific fine-tuning of the text encoder, flow-matching module, and vocoder, evaluated across seen/unseen speakers.
Repository · Demo · Paper under submission — EUSIPCO 2026
Computer-vision module for the aikon-demo platform making fine-grained, interpretable paleographic analysis accessible to historians without ML expertise: a new dataset object pairing text-line images with transcriptions, a training/inference API for morphological script-metrological analysis (following Vlachou-Efstathiou et al., ICDAR 2026), and an interactive front-end for character prototypes and diachronic evolution. Built with Mathieu Aubry's group.
Platform · Repository · Stack: Computer Vision · Python · PyTorch · web platform
Specialties: DDP/AMP distributed training · GAN & flow-matching/diffusion generation · LoRA/PEFT fine-tuning · speech representation learning · RAG & LLM orchestration · reproducible evaluation pipelines
Si ces widgets n'affichent rien (repos privés / peu de repos publics), envisage de les retirer plutôt que de laisser un encart vide — ça se voit.
| Venue | Title | Status |
|---|---|---|
| ICNLSP 2025 | Improving French Synthetic Speech Quality via SSML Prosody Control | Published — ACL Anthology |
| JEP 2026 | WavLM-Vocoder-French: Neural Waveform Resynthesis from Frozen WavLM Representations | Published — ISCA Archive |
| EUSIPCO 2026 | Europeanizing Modular Zero-Shot TTS: A Component-Level Adaptation Framework for French and German | Published — Eurasip.org |
Released models: ssml-text2breaks-fr-lora · ssml-break2ssml-fr-lora Demos: WavLM2Audio · Prosody-Control TTS · CosyVoice2-EU
| Degree | Institution |
|---|---|
| Master's degree | Université Gustave Eiffel |
| Master's degree | UVSQ (Versailles Saint-Quentin-en-Yvelines) |
| Bachelor's degree | Université Paris Descartes |
Open to research collaborations, industry partnerships, and new opportunities in controllable TTS, prosody modeling, French speech technology, multilingual voice conversion, and production-grade AI systems — especially where scientific rigor meets real deployment.
📩 Reach out via LinkedIn or email.
If you find a repository useful, a ⭐ helps visibility and supports continued maintenance.