I am a Research Engineer in Speech working within the Hi! PARIS / École Polytechnique ecosystem, contributing to research and engineering efforts in speech generation, prosody control, voice conversion, and NLP.
My work combines scientific rigor and practical implementation, with experience spanning:
- French TTS and prosody control
- SSML-based modeling
- WavLM-based speech resynthesis
- Zero-shot multilingual voice conversion
- Evaluation and reproducible research pipelines
- HPC-scale experimentation
I have developed my work in a strong academic research environment, including collaboration and scientific interactions within the École Polytechnique ecosystem.
- 🔊 Controllable French TTS with explicit prosody planning
- 🧠 SSML-based modeling for pauses, rhythm, emphasis, and timing
- 🧪 WavLM → Audio resynthesis with adversarial training and layer ablations
- 🚀 Zero-shot voice conversion using learned speech representations
- 🌍 Multilingual TTS adaptation for European low-resource languages
- 📊 Objective and perceptual evaluation for reproducible speech research
- ⚙️ Distributed training pipelines with PyTorch, DDP, AMP, and Slurm
- 📄 ICNLSP 2025 — Improving French Synthetic Speech Quality via SSML Prosody Control — published
- 📄 JEP 2026 — WavLM-Vocoder-French: Neural Waveform Resynthesis from Frozen WavLM Representations — accepted
- 📄 EUSIPCO 2026 — Europeanizing Modular Zero-Shot TTS: A Component-Level Adaptation Framework for French and German — under submission
- 🤗 Released Hugging Face models for French SSML pause prediction and break rendering
- 🎙️ Developed WavLM-based speech resynthesis with adversarial training, layer ablation studies, and open-source release
- 🛠️ Built reproducible research pipelines for:
- distributed training (DDP / AMP)
- checkpoint-based evaluation
- ablation studies
- paper-ready tables and figures
Hi! PARIS / École Polytechnique
I contribute to research and development in speech generation and language technologies, with a particular focus on:
- controllable TTS
- prosody-aware modeling
- voice conversion
- multilingual speech adaptation
- evaluation methodology
- reproducible experimentation at scale
My work has been carried out in a highly demanding research setting, including scientific interactions with senior researchers such as Éric Moulines and Reda Dehak.
Controllable French speech synthesis with explicit SSML planning for pauses, timing, and emphasis.
What this project includes
- prosody-oriented text preprocessing
- symbolic pause planning
- SSML generation
- break prediction
- evaluation utilities
- reproducible training and inference scripts
Links
- Repository
- Demo
- HF Model —
ssml-text2breaks-fr-lora - HF Model —
ssml-break2ssml-fr-lora - Paper — ICNLSP 2025
Waveform resynthesis from frozen WavLM representations for speech generation and voice conversion research.
What this project includes
- adversarial waveform reconstruction (HiFi-GAN + MPD/MSD discriminators)
- learned weighted layer fusion over WavLM-Base+ representations
- chunked inference with overlap-add
- checkpoint evaluation and layer ablation experiments
- experiment tracking for paper-ready analysis
- trained on 238h of cleaned French speech (SIWIS, M-AILABS, Common Voice)
Links
- Repository
- Demo
- HF Models
- Paper accepted at JEP 2026 (Montpellier, June 8–12)
Component-level adaptation of CosyVoice2 for European languages, with a focus on French and German.
What this project includes
- modular adaptation of a large-scale zero-shot TTS system
- language-specific fine-tuning of text encoder, flow matching module, and vocoder
- evaluation across seen and unseen speakers for French and German
- open-source release with reproducible training configs
Links
- Repository
- Demo
- Paper under submission at EUSIPCO 2026
| 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 | Accepted |
| EUSIPCO 2026 | Europeanizing Modular Zero-Shot TTS: A Component-Level Adaptation Framework for French and German | Under submission |
- training and evaluation pipelines
- reproducible experiment configurations
- analysis scripts for ablations
- paper-ready figures and tables
- Text-to-Speech (TTS)
- prosody modeling
- SSML control
- pause prediction
- voice conversion
- multilingual speech adaptation
- speech evaluation
- segmentation and alignment
- chunked inference and overlap-add reconstruction
- PyTorch
- distributed training (DDP)
- mixed precision (AMP)
- GAN training
- speech representation learning
- flow matching / diffusion-based generation
- Slurm / HPC workflows
- experiment reproducibility
- checkpoint management
- configuration-driven training
- evaluation pipelines
- LaTeX-ready reporting and analysis
| Degree | Institution |
|---|---|
| Master's degree | Université Gustave Eiffel |
| Master's degree | UVSQ (Versailles Saint-Quentin-en-Yvelines) |
| Bachelor's degree | Paris Descartes University |
I am open to research collaborations and industry partnerships in:
- controllable TTS
- prosody modeling
- French speech technology
- multilingual voice conversion
- evaluation and reproducibility for speech systems
I am especially interested in projects where scientific rigor, data confidentiality, and engineering quality matter.
If you find my repositories useful for research or development, consider giving them a star on GitHub. It helps increase visibility and supports continued maintenance and improvement.