معرفی
Simon Leglaive is a Professor at the Rennes Institute of Electronics and Telecommunications, leading research in signal processing, machine learning, and audio/speech technologies. His work focuses on advancing methods for speech enhancement, audio source separation, and generative models for audiovisual data. He has contributed extensively to techniques involving variational autoencoders (VAEs), dynamical systems, and probabilistic modeling in noisy and reverberant environments.
His research interests span signal processing fundamentals, machine learning applications to audio, and interdisciplinary projects combining speech, audio, and computer vision. Notable contributions include frameworks for unsupervised domain adaptation in speech enhancement, advanced source separation under reverberation, and human motion analysis using latent space modeling. Recent work emphasizes generative models for speech emotion recognition and robust audiovisual fusion.
Key technical trends in his publications include:
- Integration of deep learning with classical signal processing methods
- Probabilistic modeling of reverberant audio mixtures
- Unsupervised learning for speech/noise separation
- Applications of vector quantization and masked autoencoders
He has collaborated extensively with researchers from institutions like INSA Rennes and Sorbonne University, contributing to international challenges like the CHiME-7 speech enhancement task. His laboratory develops open-source tools and benchmarks for speech processing research.
