Axel Roebel is the Head of the Analysis and Synthesis of Sound (AS) team at IRCAM (STMS), where he has held the rank of Senior Research Scientist (Directeur de Recherche) since 2017. He graduated from the University of Hanover in Electrical Engineering (1990) and earned his PhD in Computer Science from the Technical University of Berlin in 1993. In 1994, he joined the German National Research Center for Information Technology (GMD-First) in Berlin, where he conducted research on the use of artificial neural networks for the adaptive modeling of signals generated by non-linear dynamical systems. In 1996, he was appointed Assistant Professor for Digital Signal Processing in the Department of Communication Sciences at the Technical University of Berlin. In 2000, he was awarded a research fellowship at CCRMA (Stanford University), where he began investigating adaptive sinusoidal modeling. Arriving at IRCAM in 2000 within the Analysis/Synthesis team, Axel Roebel has developed state-of-the-art algorithms for the analysis and transformation of speech and music. He is the author of numerous software libraries resulting from his research, such as SuperVP, an analysis and synthesis engine integrated into many professional audio tools. He was appointed Head of the AS team in 2011 and received his Habilitation (HDR) from Sorbonne University in 2013. His current research focuses on the development of deep learning techniques for music and voice processing. This includes neural vocoders, the exploration of signal representation and manipulation within latent spaces, and the study of disentanglement strategies in these spaces.

Research topics

Voice processing

• speech analysis (F0, voiced/unvoiced, glottal source)
• singing synthesis
• speech transformation - (shape invariant phase vocoder, extended source-filter speech models (PaN), neural vocoder)
• singing voice separation
• deep learning-based speech analysis, processing, and transformation,
• neural vocoder.

Music

• high-quality signal transformation based on the phase vocoder representation
• additive signal models using advanced algorithms for the analysis and representation of non-stationary signals and the development of Pm2, IRCAMs software for sinusoidal analysis/synthesis.
• structured signal models and perceptually pertinent signal descriptors (fundamental frequency, spectral envelope, ...)
• signal decomposition
• polyphonic f0 estimation

Development activities

• Multi-band Excited WaveNet Neural Vocoder (MBExWN)
• ISiS: singing synthesis software written in Python.
• as_pysrc: Python packages for signal processing
• Deep learning-based signal processing in Tensorflow
• SuperVP: an extended phase vocoder software allowing high-quality transformations of music and speech signals, and implementing new techniques for spectral envelope estimation and transformation. SuperVP is a cross-platform library that is available in the form of a command-line application (SuperVP), which is used in AudioSculpt and OpenMusic, as well as in the form of a real-time signal transformation module, which is used in Max/MSP and SuperVP-TRaX.
• VoiceForger: real-time voice transformation library based on SuperVP.
• Pm2 library and application for analysis/synthesis using advanced sinusoidal signal models
• MatMTL: a Matlab compatible c++ template library
• LibFFT: a support library for cross-platform vectorized FFT calculation

Email : Axel.Roebel (at) ircam.fr

Team : Analyse et synthèse des sons