Theo Lepage
Ph.D. student in Machine Learning → learning robust representations for speaker & language recognition
Education
Sorbonne Université (Ph.D. in Artificial Intelligence)
icon-mapParis, France
icon-calendarNov. 2022 - Nov. 2025
  • Conducting research related to "Learning speech and speaker representations for robust speaker and language recognition"
  • Supported by French ANR 'APATE' project (Forensic Deepfakes Detection Toolbox)
  • Supervised by Dr. Réda Dehak and Pr. Thierry Géraud (LRE-EPITA)
icon-mapParis, France
icon-calendarSep. 2017 - Sep. 2022
  • Signal processing and machine learning (IMAGE major) + scientific research specialization (RDI major)
Experience
Research Scientist Intern at Siemens Healthineers
icon-mapPrinceton, USA
icon-calendarFeb. 2022 - Sep. 2022
  • Focused on state-of-the-art deep learning models for MR images enhancement (denoising and super-resolution)
  • Designed a CNN architecture that leverages the attention mechanism of Vision Transformers and recovers more details compared to the solution being used in the product
icon-shs
Research Student at LRE
icon-mapParis, France
icon-calendarJan. 2020 - Jan. 2022
  • Worked on self-supervised methods applied to speaker and language recognition while doing monthly "lightning" talks about my progress (supervised by Dr. Réda Dehak)
  • Developed a label-efficient non-contrastive speaker verification model that outperforms its supervised counterpart when fine-tuned with only 2% of labeled data
  • Our work led to a publication and an oral presentation at INTERSPEECH 2022 (one of the top conferences in the field)
icon-lre
Software Developer Intern at CNRS
icon-mapParis, France
icon-calendarSep. 2020 - Jan. 2021
  • Contributed to a real-time digital holography software (C++ / CUDA) used for retinal blood flow analysis in a medical setting
  • Our work resulted in a 20x (500 to 10,000 FPS) speedup which improved substantially output images contrast and quality
  • Our refactoring and the addition of unit tests improved the stability and allowed the project to become open source
  • Founding member of the association 'Digital Holography' created to sustain the development of the software
logo-cnrs
Teaching Assistant at EPITA
icon-mapParis, France
icon-calendarSep. 2019 - Sep. 2020
  • Taught Unix concepts as well as C and Rust programming languages to undergraduates through weekly graded practicals
icon-epita
Publications
Odyssey 2024: The Speaker and Language Recognition Workshop
Theo Lepage, and Reda Dehak
Projects
Framework for training and evaluating self-supervised learning methods for speaker verification.
A tiny deep neural network framework developed from scratch in C++ and CUDA.
Tiny automatic differentiation (autodiff) engine for NumPy tensors implemented in Python.
An Optical Character Recognition software based on a simple neural network created from scratch in C.
Skills and interests
Programming
C
C++
C#
Java
Python
PHP
JS
Bash
Certificates
Driving license
Sailing instructor diploma
Languages
English (TOEIC 905)
French (native)
Data Science
PyTorch
TensorFlow
Scikit-learn
NumPy
Pandas
Passions and interests
icon-scienceScience and AI
icon-roboticsRobotics
icon-waveSailing & windsurfing