Theo Lepage
Ph.D. in AI ▹ Self-Supervised Learning for Speaker Recognition|Open to Research Scientist / Engineer Roles
Paris, France
theolepage.com
theo@theolepage.com
Theo LepageExperience
Ph.D. Researcher at EPITA Research Laboratory (LRE)
Paris, France
Nov. 2022 - Feb. 2026- Published 8 papers at top peer-reviewed venues (IEEE TASLP, Speech Communication, Interspeech)
- Proposed novel methods for self-supervised speaker recognition → SOTA performance (1.06% EER on VoxCeleb1-O)
- Introduced a latent-space positive sampling strategy (SSPS) → reducing intra-speaker variability (-58% EER for SimCLR)
- Developed and maintained "sslsv", an open-source PyTorch toolkit for self-supervised speaker verification
- Teaching assistant at EPITA for "Intro to Deep Neural Networks" and "Python for Data Science" (2023-2025)
Research Scientist Intern at Siemens Healthineers
Princeton, USAFeb. 2022 - Sep. 2022- Developed deep learning models for MR image enhancement (denoising and super-resolution)
- Designed a CNN architecture with attention mechanisms, improving reconstruction quality over existing solutions
Software Developer Intern at CNRS
Paris, FranceSep. 2020 - Jan. 2021- Contributed to a real-time digital holography software (C++/CUDA) for retinal blood flow analysis
- Achieved a 20× speedup (500 → 10,000 FPS), significantly improving image quality and contrast
- Improved software reliability through refactoring and unit testing, enabling open-source release
- Co-founded the "Digital Holography" association to sustain ongoing development
Education
Sorbonne Université (Ph.D. in Artificial Intelligence)
Paris, FranceNov. 2022 - Feb. 2026- Thesis: "Self-Supervised Learning for Speaker Recognition" (theolepage.com/phd_thesis)
- Supervised by R. Dehak (LRE-EPITA) and T. Géraud (LRE-EPITA)
École Pour l'Informatique et les Techniques Avancées - EPITA (M.Eng. in Computer Science)
Paris, FranceSep. 2017 - Sep. 2022- Major: IMAGE (Machine Learning, Deep Learning, Signal Processing, Probability & Statistics, Linear Algebra, Optimization)
- Specialization: Research (RDI) on Speaker Recognition (supervised by R. Dehak) → publication at Interspeech 2022
Publications
Speech Communication, 2026•Theo Lepage and Reda Dehak
Interspeech, 2025•Theo Lepage and Reda Dehak
IEEE TASLP, 2025•Theo Lepage and Reda Dehak
ASVspoof Workshop, 2024•Theophile Stourbe, Victor Miara, Theo Lepage, and Reda Dehak
Towards Supervised Performance on Speaker Verification with SSL by Leveraging Large-Scale ASR ModelsInterspeech, 2024•Victor Miara, Theo Lepage, and Reda Dehak
Odyssey Workshop, 2024•Theo Lepage and Reda Dehak
Interspeech, 2023•Theo Lepage and Reda Dehak
Interspeech, 2022•Theo Lepage and Reda Dehak
Projects
Deep learning toolkit based on PyTorch for training and evaluating self-supervised models for speaker verification and other speaker-related downstream tasks.
Self-supervised iterative pseudo-labeling framework to fine-tune a speech foundation model (WavLM) from a DINO-based model for speaker verification, achieving SOTA performance.
Skills & Interests
Programming
Python
C
C++
JavaScript
Bash
Data Science
PyTorch
TensorFlow
Scikit-learn
NumPy
Pandas
Languages
English (TOEIC 905)
French (native)
Passions and interests
Science and AI
Robotics
Sailing & windsurfingAn up-to-date version of this document is available at theolepage.com/resume.