Research topics
Speech/audio processing: Time-frequency analysis, spectral modeling, automatic classification, synthesis. Development of new techniques for modeling the time-trajectories of spectral parameters. Analysis and separation of harmonic and noise components. Application to speech/audio coding, high-quality synthesis/transformation, and watermarking. Deep-learning based speech/audio signals modeling, classification and transformation/synthesis. 7 past PhD Thesis (co)supervisions. Collaborations: KUL, ENIT (Tunis), LaBRI (Bordeaux), LORIA (Nancy), IRCAM (Paris), Centrale Supélec (Rennes), Arturia (Meylan), IFSTTAR (Lille), Orange Labs (Rennes).
Multimodal speech processing: Development of methods and algorithms for speech processing exploiting speech multimodality (acoustic, visual, articulatory, and gestural signals). Application to multimodal telecommunication/HMI systems and voice technologies: noise reduction, source separation, vocal activity detection, joint audiovisual coding, cued speech analysis and modeling, speaker adaptation, articulatory-to-acoustic and acoustic-to-articulatory mapping, etc. 5 past PhD thesis (co)supervisions. Collaborations: LMC Grenoble, Cardiff University, ParisTech Telecom, FTR&D, LIG Grenoble, CHU Grenoble, INSERM, CEA/Clinatec Grenoble.
Audio source separation / watermarking: Extraction of the different instruments and voices from mixed music signals. Application to active listening of music: separate manipulation and transformation, remixing. Development of a new approach, referred to as informed source separation, based on the embedding of source side-information into the mix signals using high-capacity watermarking techniques. 2 past PhD thesis (co)supervisions. Collaborations: LaBRI (Bordeaux), ParisTech Telecom, Institut Langevin (Paris).
Audio-visual scene analysis for Human-Robot Interaction: Natural audio source detection, speech enhancement in noise, speaker diarization, localization, tracking, separation, and identification for a companion humanoid robot (Nao and iCub). Fusion of audio and visual information for natural scene analysis. 5 past PhD thesis (co)supervisions. Collaborations: INRIA Rhône-Alpes, Bar-Ilan Univ. Israel, Technion IoT Israel, Centrale Supélec (Rennes).