Sound Analysis-Synthesis

The Sound Analysis-Synthesis team conducts research and development in the analysis, transformation, and synthesis of sound signals.

© Philippe Barbosa

Sound analysis encompasses methods that enable the extraction or automatic structuring of various types of information from the signal, such as the fundamental frequency or spectral evolutions determining the pitch and timbre of the perceived sound.


Information that is not strictly musical is also taken into account and is relevant to fields such as industrial acoustics, sound design, and multimedia. One notable example is the automatic indexing of sound recordings. The methods employed rely on signal processing, statistical analysis, machine learning and deep learning techniques, pattern recognition, as well as knowledge of auditory perception.


Sound transformation and synthesis techniques are initially developed to meet the needs of musicians in creating new sounds and new musical forms. This work also finds numerous applications in fields such as cinema, video games, navigation assistance, and more broadly virtual reality. Typical examples include computer-based singing synthesis and the creation of transformed and augmented voices and instruments. Research in analysis and synthesis relies on the design, on the one hand, of signal models that make it possible to describe and reproduce sound production using signal-processing methods and, on the other hand, deep learning models aimed at capturing correlations between sounds and the targeted characteristics. These models are implemented as software for computers (macOS, Windows, and Linux), specifically designed for professional or non-professional users: musicians, sound engineers, acousticians, and enthusiasts.


Axel Roebel

Responsable d'équipe

Nicolas Obin

Chercheur

Remi Mignot

Chercheur

Frederic Cornu

Ingénieur

Guillaume Doras

Chargé de R&D

Mahdi Lakbar

Chargé(e) de recherche et de développement

Simon Rouard

Doctorant

Téo Guichoux

Doctorant

Antoine Salmona

Post-doctorant

Emilio PICARD

Stagiaire

  • Sound modeling and deep learning
  • Sound characterization
  • Analysis, characterization, and transformation of musical recordings
  • Analysis, transformation, and synthesis of the voice; speaker conversion
  • Sound scene analysis

Signal processing, deep learning techniques, statistics, information theory, numerical analysis, modeling.

Related Projects

See all projects

PostGenAI@Paris

Projet ayant pour ambition de renforcer la stratégie française en intelligence artificielle en créant un pôle d’excellence international spécialisé dans l’IA post-générative

Dates : January 2025 to December 2029

EVA

Explicite Voice Attributes

Dates : October 2023 to January 2028

DeTOX

Lutte contre les vidéos hyper-truquées de personnalités françaises

Dates : January 2023 to December 2025

Related Software

Software

RAVE

RAVE is a deep learning tool for real-time audio generation and transformation via neural models, providing fast, high-quality synthesis in Max and Pd using the nn~ decoder.

Free software

Sound design and processing

Software

Partiels

Partiels is an analysis application and a collection of plug-ins that allow you to explore the content and characteristics of sounds.

Free software

Sound design and processing

Software

nn~

A max / Pd external for real-time AI audio processing.

Free software

Sound design and processing

With the supervision of:

IrcamSorbonne UniversityCNRSMinistry of Culture