I'm Theo Lepage. I am a Research Scientist at 
pyannoteAI, and I hold a Ph.D. in Artificial Intelligence from 
Sorbonne Université. My research focuses on AI/ML for Speech Processing, notably speaker recognition and speech anti-spoofing through self-supervised learning. I aim to develop technologies capable of interpreting all aspects of human conversations.
Experience
Advancing state-of-the-art speaker diarization & conversational intelligence through deep learning research in speaker recognition
Developed deep learning models for end-to-end MR image enhancement (denoising & super-resolution)
Contributed to real-time holography software for retinal blood flow analysis → 20× input image throughput
Education
Thesis: Self-Supervised Learning for Speaker Recognition • Supervised by R. Dehak and T. Géraud @ LRE-EPITA • Proposed self-supervised speaker verification methods achieving SOTA on VoxCeleb • Published 8 papers in leading speech venues (Interspeech, IEEE TASLP, ...)
Major: AI/ML for Computer Vision • Research student @ LRDE • Teaching assistant • Exchange semester @ CSUMB • International section
Ph.D. Thesis
My Ph.D. thesis focused on self-supervised speaker verification, with the objective of reaching supervised performance while reducing reliance on labeled data. The main contributions were SSPS → a novel latent-space positive sampling (-58% EER for SimCLR), and DINO-WavLM → an iterative pseudo-labeling approach to leverage WavLM without speaker labels (SOTA at 1.06% EER on VoxCeleb1-O). This research is open-sourced as sslsv, a PyTorch toolkit for self-supervised speaker verification.
Through this work, I also explored speech anti-spoofing and deepfake detection. Our WavLM-based countermeasure, submitted at the ASVspoof 2024 challenge, placed 3rd (Track 1, Open).
